JP5219499B2 - Wind noise reduction device - Google Patents

Description

  The present invention relates to a wind noise reduction device and a wind noise reduction method for reducing wind noise contained in an input acoustic signal, and also relates to a recording device and an imaging device using the wind noise reduction device.

  In a recording apparatus equipped with a microphone, wind noise is mixed into an acoustic signal when wind strikes the microphone. This wind noise is generated when a wind pressure is applied to the diaphragm of the microphone, and is preferably removed because it becomes noise for the original acoustic signal.

  The frequency band in which wind noise exists is relatively low, and normally wind noise is concentrated in a band of about 300 Hz or less. Utilizing such characteristics, the conventional wind noise reduction apparatus has been designed to reduce wind noise centering on low-band signals. Generally, as shown in FIG. 11, a high-pass filter (HPF) and a low-pass filter (LPF) are used to separate an input acoustic signal into a low-band component and a higher-band component to reduce a low-band signal. After that (or after cutting), the method of adding both together again is taken.

  There has also been proposed a wind noise reduction device having a portion for determining the presence or absence of wind noise. The presence / absence of wind noise generally uses the feature that “wind noise has no cross-correlation between left and right channel signals”. Specifically, the cross-correlation is obtained between the left and right channel signals forming the input acoustic signal, and the input acoustic signal contains wind noise when the correlation value representing the cross-correlation is a certain threshold value or less. to decide. In addition to simply determining the presence or absence of wind noise, the obtained correlation value is also used as an index representing the strength of wind noise. For example, a method of varying the degree of reduction of the low-band signal according to the correlation value has been proposed (see, for example, Patent Document 1 below).

Japanese Patent Laid-Open No. 11-69480

  The low band includes a frequency band of wind noise and is greatly affected by wind noise, but the low band includes important elements of sound. In particular, with regard to human voice, the pitch (pitch frequency) of the voice is 90 to 160 Hz for males and 230 to 370 Hz for females, and very important elements for determining sound quality are included in the low band. Yes. The pitch is a fundamental frequency of a signal due to vocal cord vibration. If a band component including such an important element is simply reduced or cut, the signal component element different from the wind noise is reduced or cut, resulting in a distorted sound. In the case of a human voice, the voice becomes smaller or the voice color changes.

  Also, if wind noise reduction measures are taken only in the low band and no wind noise reduction measures are taken in the other bands, wind noise with a relatively high frequency (sounds like a clog) will remain and the user will feel uncomfortable. Remember.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a wind noise reduction device and a wind noise reduction method that generate little distortion and have a high wind noise reduction effect. It is another object of the present invention to provide a recording device and an imaging device using the wind noise reduction device.

  In order to achieve the above object, a wind noise reduction device according to the present invention is a wind noise reduction device that generates a corrected acoustic signal in which wind noise is reduced from an input acoustic signal. When the second band is a predetermined band having a frequency higher than that of the first band, the wind noise reduction device converts the sound signal in the band having a frequency higher than the first band included in the input acoustic signal. Based on the acoustic signal generated by the signal generating means, and having a signal generating means for generating the acoustic signal of the first band different from the acoustic signal of the first band included in the input acoustic signal. A first correcting means for generating a first corrected acoustic signal, and a second acoustic signal in the second band in which the wind noise is reduced by reducing the signal level of the second acoustic signal in the input acoustic signal. Corrected acoustic signal A second correction means for generating, characterized by comprising a correction sound signal output means for outputting the corrected sound signal based on the first and second correction acoustic signal.

  Acoustic signal generation processing (for example, signal restoration processing) is performed on the first low frequency band containing important sound elements to reduce wind noise while suppressing sound distortion. Plan. On the other hand, wind noise is reduced by a signal reduction process for the second band on the higher frequency side than the first band, where distortion is less likely to occur due to signal reduction. Thereby, the high wind noise reduction effect is acquired, suppressing generation | occurrence | production of distortion.

  Specifically, for example, the first correcting unit generates the first corrected acoustic signal based on the acoustic signal in the first band included in the input acoustic signal and the acoustic signal generated by the signal generating unit. To do.

  More specifically, for example, the input acoustic signal includes a plurality of channel signals, and the wind noise reduction device further includes a predetermined band component of each channel signal including the wind noise band between different channel signals. Wind noise determination means for determining the degree of influence of the wind noise on the input acoustic signal based on the cross-correlation in the above, the first correction means according to the determination result of the wind noise determination means A corrected acoustic signal is generated.

  Further, for example, the input acoustic signal is composed of a plurality of channel signals, and the wind noise reduction device further performs cross-correlation between different channel signals of predetermined band components of each channel signal including the wind noise band. Based on the wind noise determining means for determining the degree of influence of the wind noise on the input acoustic signal, and the second correcting means determines the second corrected acoustic signal according to the determination result of the wind noise determining means. Generate.

  Alternatively, for example, the input acoustic signal is given to the wind noise reduction device as a signal on the frequency axis, and the input acoustic signal is composed of a plurality of channel signals, and the second correcting means is The second corrected acoustic signal on the frequency axis is generated by dividing the two bands into a plurality of element bands and reducing the signal level of the acoustic signal of each element band, and for each of the plurality of element bands, The cross-correlation of the acoustic signal of the element band between different channel signals is obtained, and the degree of reduction of the signal level is determined for each element band based on each cross-correlation.

  Thereby, the wind noise reduction effect with respect to the second band can be obtained without generating distortion unnecessarily.

  Also, for example, the input acoustic signal is given to the wind noise reduction device as a signal on the time axis, and the input acoustic signal is composed of a plurality of channel signals and is generated by the first correction means. The acoustic signal is a signal on a time axis, and the audio signal correction apparatus further includes an extraction unit that extracts a component of a predetermined band that does not include the first band and includes the second band from the input acoustic signal. A time frequency conversion means for converting a signal format of a synthesized signal of the first corrected acoustic signal and the extraction signal of the extraction means from a time axis to a frequency axis, and the second correction means includes a frequency axis The second corrected acoustic signal on the frequency axis is generated by reducing the signal level of the second band acoustic signal in the synthesized signal above, and the corrected acoustic signal output means is provided by the second correcting means. The corrected acoustic signal on the frequency axis is output based on the second corrected acoustic signal on the frequency axis and the acoustic signal including the first corrected acoustic signal on the frequency axis obtained from the time-frequency conversion means. To do.

  With this configuration, the wind noise reduction device can be easily incorporated into the encoder.

  And, for example, the second correction means divides the second band of the synthesized signal on the frequency axis into a plurality of element bands, and reduces the signal level of the acoustic signal in each element band, thereby reducing the first level on the frequency axis. 2 corrected acoustic signals are generated, cross-correlation of the acoustic signals of the element bands between different channel signals is obtained for each of the plurality of element bands, and the signal for each element band is obtained based on each cross-correlation Determine the level of reduction.

  Further, for example, the input acoustic signal is composed of a plurality of channel signals, and the second correction means focuses on all or a part of the second band, and the plurality of channel signals included in the input acoustic signal. The signal level of the acoustic signal in the bandwidth of interest in a channel that is relatively affected by the wind noise is reduced by averaging the acoustic signal in the bandwidth of interest in minutes, and the signal obtained by this averaging is reduced. The second corrected acoustic signal is generated.

  Further, for example, the input acoustic signal is composed of a plurality of channel signals, and the second correction means focuses on all or a part of the second band, and the plurality of channel signals included in the input acoustic signal. Minute acoustic signal having the minimum signal level is specified as the minimum acoustic signal and other acoustic signals are specified as non-minimum acoustic signals, and the non-minimum acoustic signal is determined by the minimum acoustic signal. By replacing, the signal level of the acoustic signal in the band of interest in a channel that is relatively affected by the wind noise is reduced, and the second corrected acoustic signal is generated from the signal obtained by this replacement.

  In order to achieve the above object, a recording apparatus according to the present invention includes the wind noise reduction device and a microphone for generating the input acoustic signal for the wind noise reduction device. .

  In order to achieve the above object, an imaging apparatus according to the present invention includes the wind noise reduction apparatus, a microphone for generating the input acoustic signal for the wind noise reduction apparatus, and an imaging means for acquiring an image. And.

  In order to achieve the above object, a wind noise reduction method according to the present invention is a wind noise reduction method for generating a corrected acoustic signal in which wind noise is reduced from an input acoustic signal. When the second band is a predetermined band having a frequency higher than that of the first band and the first band, the wind noise reducing method includes sound in a band higher in frequency than the first band included in the input acoustic signal. Based on the signal, a signal generation step of generating an acoustic signal of the first band different from the acoustic signal of the first band included in the input acoustic signal, and the acoustic signal generated in the signal generation step A first correction step for generating a first corrected acoustic signal, and a second-band acoustic signal in which the wind noise is reduced by reducing a signal level of the second-band acoustic signal of the input acoustic signal. That a second correction step of generating a second correction acoustic signals, comprises a, and generates the corrected sound signal based on the first and second correction acoustic signal.

  According to the present invention, it is possible to provide a wind noise reduction device, a wind noise reduction method, a recording device, and an imaging device that generate little distortion and have a high wind noise reduction effect.

  The significance or effect of the present invention will become more apparent from the following description of embodiments. However, the following embodiment is merely one embodiment of the present invention, and the meaning of the term of the present invention or each constituent element is not limited to that described in the following embodiment. .

  Hereinafter, embodiments of the present invention will be specifically described with reference to the drawings. In each of the drawings to be referred to, the same part is denoted by the same reference numeral, and redundant description regarding the same part is omitted in principle. The first to fifth embodiments will be described later. First, matters common to each embodiment or items referred to in each embodiment will be described.

  FIG. 1 is an external perspective view of an imaging apparatus 1 according to an embodiment of the present invention. The imaging device 1 is a digital video camera that can also record audio. A microphone MIC1 is provided on the left side of the casing of the imaging apparatus 1, and a microphone MIC2 is provided on the right side of the casing of the imaging apparatus 1. The microphone MIC1 picks up the sound coming from the left direction of the imaging device 1, and the microphone MIC2 picks up the sound coming from the right direction of the imaging device 1, whereby the microphones MIC1 and MIC2 form a stereo microphone. Although different from the arrangement position shown in FIG. 1, the microphones MIC1 and MIC2 can be installed close to each other, for example, on the back surface of the plate-like housing in which the display display is fitted (the surface on the opposite side of the display display). .

  FIG. 2 is a schematic block diagram showing the electrical configuration of the imaging apparatus 1. The imaging device 1 includes an imaging unit 2, a video signal processing unit 3, an audio signal processing unit 4, and a recording medium 5 in addition to the microphones MIC 1 and MIC 2. Although not shown, the imaging apparatus 1 includes an operation unit including a shutter button and a recording button, a display display, a speaker, a CPU (Central Processing Unit), and the like.

  The imaging unit 2 includes an optical system and an imaging device such as a charge coupled device (CCD) or a complementary metal oxide semiconductor (CMOS) image sensor, and converts an optical image incident through the optical system into an electrical signal. An image represented by the electrical signal is acquired. The video signal processing unit 3 generates a video signal representing an acquired image of the imaging unit 2 based on the electrical signal. The video signal is recorded on a recording medium 5 such as a memory card or an optical disk in accordance with an operation on an operation unit (not shown) provided in the imaging apparatus 1.

  Each of the microphones MIC1 and MIC2 converts the sound collected by itself into an analog electric signal and outputs it. Each output signal of the microphones MIC1 and MIC2 is converted into a digital signal by an A / D converter (not shown) provided in the audio signal processing unit 4, and the audio signal processing unit 4 converts the digital signal into a desired process. Apply. A signal obtained through this processing is recorded on the recording medium 5 in accordance with an operation on an operation unit (not shown) provided in the imaging apparatus 1.

  The microphones MIC1 and MIC2 each have a diaphragm (not shown) as a vibrating body. Each diaphragm vibrates due to air vibration caused by sound waves, but also vibrates due to wind pressure acting on the diaphragm. Therefore, when the sound wave and the wind pressure act on the diaphragm, the diaphragm vibrates according to the sound wave and the wind pressure. Each of the microphones MIC1 and MIC2 converts the vibration of each diaphragm into an electrical signal and outputs it. Of the output signal of the microphone, noise (noise component) derived from wind pressure is called wind noise. Wind noise is not noise that arrives at the diaphragm as sound waves.

  The audio signal processing unit 4 includes a wind noise reduction unit 6. The wind noise reduction unit 6 reduces wind noise from the input signal of the wind noise reduction unit 6 based on the output signal of each microphone, and outputs an acoustic signal with reduced wind noise as an output signal.

  The frequency band in which wind noise exists is relatively low, and normally wind noise is concentrated in a band of about 300 Hz or less. Therefore, the wind noise reduction unit 6 treats 300 Hz as a boundary, treats a frequency band smaller than 300 Hz (Hertz) as a “low band”, and performs processing for reducing wind noise on the low band. However, although the intensity is relatively small, wind noise exists in a frequency band of 300 Hz or higher and close to a low band. Therefore, the wind noise reduction unit 6 further handles the frequency band of 300 Hz or more by dividing it into a medium band and a high band, and performs processing for reducing the wind noise for the medium band. As a specific numerical example, a frequency band of 300 Hz or more and smaller than 1.5 kHz (kilohertz) is treated as a “medium band”, and a frequency band of 1.5 kHz or more is treated as a “high band”.

  The low band includes a frequency band of wind noise and is greatly affected by wind noise, but the low band includes important elements of sound. In particular, with regard to human voice, the pitch (pitch frequency) of the voice is 90 to 160 Hz for males and 230 to 370 Hz for females, and very important elements for determining sound quality are included in the low band. Yes. The pitch is a fundamental frequency of a signal due to vocal cord vibration. If a band component including such an important element is simply reduced or cut, the signal component element different from the wind noise is reduced or cut, resulting in a distorted sound. In the case of a human voice, the voice becomes smaller or the voice color changes.

  Therefore, the wind noise reduction unit 6 divides the process for reducing the wind noise into two stages, and applies each process to different bands. Of the two stages, one process is a signal restoration process that restores a signal that does not include wind noise, and the other process is a signal reduction process that reduces wind noise by reducing the signal level.

  The signal restoration process is applied to a low-band signal. Since the low band includes important elements of sound along with strong wind noise, noise reduction is attempted by restoring a signal that does not include wind noise, rather than reducing the signal level. If the signal restoration process is performed, it is not necessary to reduce the signal level, so that distortion of sound is less likely to occur.

  The signal reduction process is applied to a medium band signal. The effect of wind noise on the mid-band is small, but wind noise at a relatively high frequency (coro And the user feels uncomfortable. However, since the effects of wind noise are small, it is assumed that the sound distortion due to signal reduction is small, and even if attention is paid to sound elements, the middle band is a band where harmonics of pitch exist, so signal reduction can be performed. The lower the band, the less affected by distortion. Therefore, as described above, signal reduction processing is applied to the mid-band signal.

  Although it is conceivable to apply the signal restoration processing to the middle band as well, in order to restore the middle band signal that does not include wind noise, harmonic components in the high band signal are required. Since such harmonic components are weak, it is difficult to restore them satisfactorily. Therefore, signal reduction processing is suitable for a medium-band signal.

  Either the signal restoration process or the signal reduction process may be performed first, or each may be performed in parallel. Each of the signal restoration processing and the signal reduction processing can be performed on the time axis and the frequency axis.

  Further, a wind noise determination unit that determines the presence or absence and strength of wind noise may be provided. The wind noise determination unit determines, for example, the presence or intensity of wind noise by obtaining a cross-correlation between the left and right channels, and the determination result is used for signal restoration processing and / or signal reduction processing. One wind noise determination unit may be shared between the signal restoration process and the signal reduction process, or two wind noise determination units are provided, and the wind noise determination unit is independently provided for each of the signal restoration process and the signal reduction process. May be assigned. When the wind noise determination unit is independently assigned to each of the signal restoration process and the signal reduction process, it is possible to mutually use the respective determination results (detailed specific examples will be described later).

  Cross-correlation means mutual correlation between signals to be compared. In each embodiment described later, a correlation value obtained through a predetermined calculation is handled as an index indicating cross-correlation, but the cross-correlation evaluation method is not limited to this.

  Below, the 1st-5th Example is described as a specific Example in connection with the wind noise reduction part 6. FIG.

<< First Example >>
First, the first embodiment will be described. In the first embodiment, both the signal restoration process and the signal reduction process are executed on the time axis.

  FIG. 3 is an internal block diagram of the wind noise reduction unit 6a according to the first embodiment. The wind noise reduction unit 6a is used as the wind noise reduction unit 6 of FIG. The wind noise reduction part 6a is provided with each site | part referred with the codes | symbols 11-15.

  An input signal (input acoustic signal) to the wind noise reduction unit 6a is an acoustic signal on the time axis composed of a plurality of channel signals (in other words, an acoustic signal expressed in the time domain). Specifically, the audio signal processing unit 4 in FIG. 2 converts analog output signals from the microphones MIC1 and MIC2 into digital signals at a predetermined sampling frequency. Now, a channel signal in which digital signals corresponding to the output signal of the microphone MIC1 are arranged in time series is represented by L (t), and a channel signal in which digital signals corresponding to the output signal of the microphone MIC2 are arranged in time series is R ( t). A channel signal corresponding to the output signal of the microphone MIC1 is called an L signal, and a channel signal corresponding to the output signal of the microphone MIC2 is called an R signal. Then, the input signal to the wind noise reduction unit 6a in FIG. 3 includes the L signal L (t) and the R signal R (t). This input signal is corrected by the wind noise reduction unit 6a. Therefore, an input signal to the wind noise reduction unit 6a is referred to as an “original signal”, and an output signal of the wind noise reduction unit 6a is referred to as a “correction signal”. Hereinafter, the L signal L (t) and the R signal R (t) may be abbreviated as the signals L (t) and R (t), respectively.

  The signals L (t) and R (t) take positive and negative values centered on zero, and when the diaphragms of the microphones MIC1 and MIC2 do not vibrate, the values of L (t) and R (t) are zero ( However, offset and noise components are ignored), and the amplitudes of L (t) and R (t) increase as their vibration increases.

  In the first embodiment, the original signal is supplied to band pass filters (hereinafter referred to as BPF) 23 and 30, low pass filters (hereinafter referred to as LPF) 21 and 26, and high pass filter (hereinafter referred to as HPF) 14, respectively. Is input.

  The wind noise determination unit 11 includes parts referred to by reference numerals 21 and 22. The LPF 21 extracts and outputs a predetermined band component of an input signal for itself. The band extracted by the LPF 21 includes the frequency band of wind noise, and is typically the same as the “low band” described above. However, the band extracted by the LPF 21 does not have to completely match the above-mentioned “low band”. For example, a frequency band smaller than 200 Hz may be a band extracted by the LPF 21.

  By the way, each part in the wind noise reduction unit 6a individually performs necessary signal processing on each of the plurality of channel signals. That is, for example, the LPF 21 extracts and outputs a predetermined band component of the L signal L (t) and a predetermined band component of the R signal R (t). This also applies to wind noise reduction units 6b and 6c described later. However, as a matter of course, this does not apply to the part (correlation value calculation unit 22 in the present embodiment) where the correlation value is calculated by cross-correlation calculation.

  The correlation value calculation unit 22 obtains a correlation value representing a cross-correlation between channel signals output from the LPF 21, that is, a correlation value between channel signals in the band component extracted by the LPF 21. Specifically, the original signal and the acoustic signal on the time axis based on the original signal are considered by dividing them into unit intervals of a predetermined interval. As shown in FIG. 4, it is assumed that time advances in the order of the first, second, third unit intervals,..., And each unit interval includes N discrete signals (N samples). Signal). Accordingly, one unit section includes N L signals L (t) and N R signals R (t).

The correlation value calculation unit 22 calculates a correlation value K [p] for each unit section according to the following equation (1) based on the signals L (t) and R (t) output from the LPF 21. Here, p represents a unit interval number. L _i and R _i in equation (1) indicate the value of the i-th L signal L (t) and the value of the i-th R signal R (t), respectively, in the unit interval of interest. Of course, since the signals L (t) and R (t) for the correlation value calculation unit 22 are given through the LPF 21, L _i and R _i in the equation (1) are defined by the output value of the LPF 21.

  Wind noise has no cross-correlation between left and right channel signals. Therefore, the correlation value is relatively small if the wind noise contained in the original signal is relatively large, and the correlation value is relatively large if the wind noise contained in the original signal is relatively small. Accordingly, the correlation value K [p] takes a value corresponding to the wind noise intensity in the p-th unit interval. Using this, the wind noise determination unit 11 determines the degree of influence of wind noise on each unit section based on the correlation value calculated by the correlation value calculation unit 22. This determination result is used for processing in the signal restoration unit 12 and the signal reduction unit 13.

  The signal restoration unit 12 includes each part referred to by reference numerals 23 to 29. The signal restoration unit 12 generates a restoration signal for the low band from the middle band signal of the original signal by using the harmonic nature of the voice or instrument sound.

  Harmonicity is the property that the frequency spectrum is composed of a harmonic structure, and many voices and instrument sounds have this property. That is, in the frequency spectrum of a certain sound, if the frequency of the lowest frequency component is f0, the frequency spectrum of the sound is f0 and its overtone components f0 × 2, f0 × 3, f0 × 4, Are formed from frequency components. In this case, the frequency component of f0 is called a fundamental wave component, and the frequency components of f0 × 2, f0 × 3, f0 × 4,... Are second order, third order, fourth order,. It is called a harmonic component.

  For harmonic signals, it is known that the fundamental or low-order harmonic components can be restored from the higher-order harmonic components, and this restoration can be performed using nonlinear processing such as squaring, full-wave rectification, and half-wave rectification. It is known that processing can be used (for example, JP-A-8-130494, JP-A-8-278800, JP-A-9-55778).

  The signal restoration unit 12 in FIG. 3 can also generate a restoration signal using any known method. As a specific example, in the signal restoration unit 12, a restoration signal is generated by each part referred to by reference numerals 23 to 25. Each part will be described.

  The BPF 23 extracts and outputs a predetermined band component of an input signal for itself. In order to restore a low-band signal, the band extracted by the BPF 23 is the same as the “medium band” described above. However, the band extracted by the BPF 23 does not have to completely match the above “medium band”.

  The nonlinear processing unit 24 performs nonlinear processing on the signal that has passed through the BPF 23 (the signal extracted by the BPF 23). This non-linear processing is square processing, full-wave rectification (absolute value processing), half-wave rectification, or the like. When square processing is used, the nonlinear processing unit 24 squares and outputs the signal that has passed through the BPF 23. When a human voice is picked up by the microphones MIC1 and MIC2, the signal that has passed through the BPF 23 contains harmonic components of the pitch signal of the voice. By squaring the passed signal, the frequency of each harmonic component is obtained. A signal having a frequency corresponding to the difference and the sum is generated. That is, harmonic components (fundamental wave components or harmonic components) are also generated on the low frequency side and the high frequency side of the passband of the BPF 23 by the square process. When the square process is used, the amplitude of the generated harmonic component is the square of that of the harmonic component to be originally obtained. Therefore, when using the square process, the nonlinear processing unit 24 performs normalization on the square signal obtained by squaring the signal that has passed through the BPF 23, and outputs a square signal whose amplitude is adjusted thereby. Like that.

  The same applies when full-wave rectification (absolute value processing) or half-wave rectification is used as the nonlinear processing. For example, when using full-wave rectification, the nonlinear processing unit 24 calculates and outputs the absolute value of the signal that has passed through the BPF 23.

  The signal restoration unit 12 uses only the low-band signal component of the restored signal. For this reason, the LPF 25 passes only the low-band component of the output signal of the nonlinear processing unit 24. The output signal of the LPF 25 is a low-band acoustic signal restored from the middle-band acoustic signal in the original signal. And since the wind band is hardly contained in the middle band that is the source of restoration, the restored low-band acoustic signal contains almost no wind noise. That is, a low-band acoustic signal in which wind noise is reduced as compared with the low-band acoustic signal of the original signal is restored by each part referred to by reference numerals 23 to 25.

  On the other hand, the signal restoration unit 12 prepares an original low-band signal using the LPF 26. That is, the LPF 26 that passes only the low-band component of the input signal to itself is used, and only the low-band component of the original signal is output from the LPF 26.

The multipliers 27 and 28 and the adder 29 perform weighted addition of the output signal values of the LPFs 25 and 26 according to the correlation value calculated by the correlation value calculation unit 22, thereby outputting the output signal (first signal) of the signal restoration unit 12. 1 corrected acoustic signal) is generated. When the output signal value of the LPF 26 in the p-th unit interval is expressed by LPF_OUT _O (t) and the output signal value of the LPF 25 in the p-th unit interval is expressed by LPF_OUT _R (t), the p-th unit The output signal value OUT ₁₂ (t) of the signal restoration unit 12 corresponding to the section is expressed by the following equation (2).

  That is, when the correlation value is relatively large, it is determined that the wind noise is relatively small, so that the contribution of the original low-band signal to the output signal of the signal restoration unit 12 is increased. On the other hand, since it is determined that the wind noise is relatively large when the correlation value is relatively small, the degree of contribution of the restoration signal (the low-band signal of the restoration signal) to the output signal of the signal restoration unit 12 is increased.

As can be seen from the calculation formula of the above formula (1), the correlation value K [p] satisfies the inequality “0 ≦ K [p] ≦ 1”. For this reason, the correlation value K [p] is used as it is for the calculation of OUT ₁₂ (t), but if “0 ≦ K [p] ≦ 1” is not satisfied, Equation (2) may be changed as appropriate. Further, when the correlation value is larger than a predetermined reference threshold with respect to the focused unit section, it may be determined that there is no wind noise and the output signal of the LPF 26 may be directly used as the output signal of the signal restoration unit 12.

  The signal reduction unit 13 includes portions referred to by reference numerals 30 and 31. The BPF 30 extracts and outputs the middle band component of the input signal for itself. For each unit section, the multiplier 31 reduces the level of the signal that has passed through the BPF 30 (that is, the middle-band acoustic signal extracted from the original signal) according to the correlation value calculated by the correlation value calculator 22. The reduced signal is output as an output signal of the signal reduction unit 13. The level of a certain signal represents the amplitude (intensity) of the signal.

  At this time, if it is determined from the correlation value that the influence of wind noise is large, the level is greatly reduced, and if it is determined from the correlation value that the influence of wind noise is small, the level is not reduced so much. That is, when paying attention to the p-th unit interval, the level is reduced more greatly by increasing the reduction rate corresponding to the p-th unit interval as the correlation value K [p] decreases (conversely, As the correlation value K [p] increases, the reduction rate corresponding to the p-th unit interval is decreased). By the signal reduction by the multiplier 31, the wind noise included in the output signal (second corrected acoustic signal) of the signal reduction unit 13 is appropriately reduced.

  In addition, if the same result is led, the method of reducing is arbitrary. For example, the output value of the BPF 30 may be multiplied by the correlation value calculated by the correlation value calculation unit 22 or a coefficient corresponding to the correlation value.

  The HPF 14 passes only the high band component of the input signal to itself.

  The signal synthesis unit 15 outputs an output signal of the signal restoration unit 12 that represents a low-band acoustic signal in which wind noise has been reduced by signal restoration processing, and a signal that represents an intermediate-band acoustic signal in which wind noise has been reduced by signal reduction processing. The output signal of the reduction unit 13 and the output signal of the HPF 14 are added, and a signal obtained by this addition is output as an output signal (that is, a correction signal) of the wind noise reduction unit 6a. In the first embodiment, this correction signal is also an acoustic signal on the time axis composed of a plurality of channel signals, like the original signal.

  When the signal delay amounts in the signal restoration unit 12, the signal reduction unit 13, and the HPF 14 are different, a delay process that cancels the difference between the signal delay amounts in the signal synthesis unit 15 or in the previous stage of the signal synthesis unit 15. Then, the signal synthesizer 15 performs the addition process. The same applies to the weighted addition processing using the multipliers 27 and 28 and the adder 29. Further, it is necessary to calculate a correlation value before the signal restoration process by the signal restoration unit 12 and the signal reduction process by the signal reduction unit 13, but there is no particular limitation on the processing order of the signal restoration process and the signal reduction process.

  The audio signal processing unit 4 in FIG. 2 records a signal obtained by performing a predetermined encoding process (audio compression process) on the correction signal output from the signal synthesis unit 15 on the recording medium 5. The predetermined encoding process is, for example, AAC (Advanced Audio Coding) according to the MPEG (Moving Picture Experts Group) standard.

  In the above description, in principle, the signal processing for the L signal and the R signal is not described separately. However, as described above, each part in the wind noise reduction unit 6a is assigned to each of a plurality of channel signals. On the other hand, necessary signal processing is performed individually.

That is, the LPF 21 extracts and outputs each predetermined band component (typically a low band component) of the L signal and the R signal forming the original signal. The BPF 23 extracts and outputs each predetermined band component (typically a middle band component) of the L signal and the R signal forming the original signal. The nonlinear processing unit 24 individually performs nonlinear processing on the L signal and the R signal given through the BPF 23, and the LPF 25 passes only the low-band components of the L signal and R signal after the nonlinear processing. The LPF 26 passes only the low-band components of the L signal and R signal forming the original signal. The multipliers 27 and 28 and the adder 29 perform weighted addition processing of the L signal output from the LPF 25 and the L signal output from the LPF 26 and weight the R signal output from the LPF 25 and the R signal output from the LPF 26. Addition processing is performed.
The BPF 30 extracts and outputs each of the middle band components of the L signal and the R signal forming the original signal, and the multiplier 31 reduces the levels of the L signal and the R signal that have passed through the BPF 30 according to the correlation value. (The correlation value that determines the reduction rate is common between the L signal and the R signal).
The HPF 14 passes only the high-band components of the L signal and R signal forming the original signal. The signal synthesis unit 15 adds the L signal in the output signal of the signal restoration unit 12, the L signal in the output signal of the signal reduction unit 13, and the L signal in the output signal of the HPF 14, and also the R signal in the output signal of the signal restoration unit 12. And the R signal in the output signal of the signal reduction unit 13 and the R signal in the output signal of the HPF 14 are added to generate a correction signal.

  In addition, the wind noise determination unit 11 can be omitted from the wind noise reduction unit 6a. When the wind noise determination unit 11 is omitted, the multipliers 27 and 28 and the adder 29 perform weighted addition of the output signal values of the LPFs 25 and 26 at a constant ratio, thereby generating an output signal (first signal) of the signal restoration unit 12. Corrected acoustic signal) is generated. That is, in this case, K [p] in the above equation (2) is a fixed value. When the wind noise determination unit 11 is omitted, the multiplier 31 reduces the level of the signal that has passed through the BPF 30 at a constant reduction rate, and outputs the reduced signal as an output signal of the signal reduction unit 13. . When the wind noise determination unit 11 is omitted, the input signal to the wind noise reduction unit 6a may be a monaural signal composed of one channel signal.

  In the above example, the BPF 23, the non-linear processing unit 24, and the LPF 25 perform necessary processing individually on the L signal and the R signal, thereby generating a restoration signal for the L signal and a restoration signal for the R signal. However, a monaural signal may be generated from the L signal and R signal forming the original signal, and a monaural restoration signal may be generated based on the monaural signal. Signal monauralization can be performed at any stage in the process of generating a restored signal from the original signal. Normally, a monaural signal is generated by averaging the L signal and the R signal forming the original signal before the BPF 23, and the monaural signal is supplied to the BPF 23. The monaural restoration signal is used as a restoration signal for the L signal and also as a restoration signal for the R signal. If a monaural restoration signal is generated from a monaural signal, signal processing is only required for one channel, and the processing can be simplified. Since the sense of stereo is poor in the low band, there are few problems even if a monaural restoration signal is used. The technical matter of generating a monaural restoration signal can also be applied to other embodiments described later.

[Modification of signal reduction processing]
In the signal reduction processing described above, the multiplier 31 is used to reduce the level of the signal that has passed through the BPF 30 at a reduction rate corresponding to the correlation value K [p]. It is not limited. As an example, first and second modified signal reduction processes related to the signal reduction process of the first embodiment will be exemplified. Hereinafter, a channel corresponding to the L signal is referred to as an L channel, and a channel corresponding to the R signal is referred to as an R channel.

First, the first deformation signal reduction process will be described. In the first modified signal reduction process, the signal reduction unit 13 compares the correlation value K [p] with a predetermined threshold value K _THA . As described above, the correlation value K [p] represents the degree of wind noise influence in the p-th unit interval. On the other hand, the threshold value K _THA represents a reference influence degree to be compared with the influence degree. If the correlation value K [p] is smaller than the threshold value K _THA , it is judged that the influence of wind noise in the p-th unit interval is relatively large. If the correlation value K [p] is larger than the threshold value K _THA , It is determined that it is relatively small (the same applies to the second deformation signal reduction process).

When the correlation value K [p] is smaller than the threshold value K _THA , the signal reduction unit 13 averages the L signal and the R signal that have passed through the BPF 30, and the monaural signal obtained by this averaging is used as the signal reduction unit 13. To the signal synthesizer 15 as an output signal. Now, the signal values of the L signal and the R signal in the p-th unit interval that have passed through the BPF 30 are represented by BPF_OUT _L (t) and BPF_OUT _R (t), respectively, and are output from the signal reduction unit 13. The signal values of the L signal and the R signal in the unit interval are represented by BPF_OUT _L ′ (t) and BPF_OUT _R ′ (t), respectively. Then, when the correlation value K [p] is smaller than the threshold value K _THA, it is expressed by “BPF_OUT _L '(t) = BPF_OUT _R ′ (t) = (BPF_OUT _L (t) + BPF_OUT _R (t)) / 2”. Is output from the signal reduction unit 13.

  Since wind noise is noise that is randomly generated in each channel due to turbulence, there are channels that are strongly influenced by wind noise and channels that are weak. If the above averaging is performed, the influence of wind noise is equalized between the channels, and the noise level of a channel that is relatively affected by wind noise is reduced.

  FIG. 12 shows a conceptual diagram of signal reduction processing by this averaging. The p-th unit interval will be discussed. In the example shown in FIG. 12, the influence of wind noise is relatively large for the L channel and relatively small for the R channel. Accordingly, the signal level of the L signal that has passed through the BPF 30 is equal to that of the BPF 30. It is larger than that of the R signal that has passed. If the above averaging is performed in such a case, the wind noise components included in the L signal and the R signal that have passed through the BPF 30 are also averaged, and as a result, the signal level of the L signal among the passing signals of the BPF 30 is reduced. Is done.

On the other hand, when the correlation value K [p] is larger than the predetermined threshold value K _THA , the L signal and the R signal that have passed through the BPF 30 should be output as they are from the signal reduction unit 13 without performing the above averaging. It is good to give to the signal composition part 15 as L signal and R signal. Alternatively, signal reduction processing as in principle using the multiplier 31 may be performed. That is, when the correlation value K [p] is larger than the predetermined threshold value K _THA , the multiplier 31 is used to reduce the level of the L signal and the R signal that have passed through the BPF 30 to a reduction rate according to the correlation value K [p]. The signal obtained by this may be used as the output signal of the signal reduction unit 13.

  Next, the second deformation signal reduction process will be described. When the first modified signal reduction process is used, the wind noise component of the channel (the R channel in FIG. 12) where the influence of the wind noise is relatively small is increased by the signal reduction process, as can be understood from FIG. . In order to avoid this, in the second modified signal reduction process, the signal of the channel with the higher signal level is replaced with the signal of the channel with the lower signal level of the left and right channels.

More specifically, in the second modified signal reduction process, the signal reduction unit 13 compares the correlation value K [p] with a predetermined threshold value K _THA . When the correlation value K [p] is smaller than the threshold value K _THA , the signal having the smaller signal level of the L signal and the R signal that have passed through the BPF 30 is set as the minimum acoustic signal and the other signal is set as the non-minimum acoustic signal. Identify as signal and replace non-minimum acoustic signal with minimum acoustic signal. That is, when the correlation value K [p] is smaller than the threshold value K _THA , among the L signal and R signal that have passed through the BPF 30, for example, the R signal is specified as the minimum acoustic signal, “BPF_OUT _L ′ (t ) = BPF_OUT _R '(t) = BPF_OUT _R (t) "is output from the signal reduction unit 13.

  FIG. 13 shows a conceptual diagram of signal reduction processing by this replacement. The p-th unit interval will be discussed. In the example shown in FIG. 13, the influence of wind noise is relatively large for the L channel and relatively small for the R channel. Accordingly, the signal level of the L signal that has passed through the BPF 30 is the same as that of the BPF 30. It is larger than that of the R signal that has passed. When the above replacement is performed in such a case, the wind noise component included in the L signal that has passed through the BPF 30 is reduced (no change with respect to the R signal). As a result, the noise level of a channel that is relatively affected by wind noise can be reduced without increasing the noise level of a channel that is relatively less affected by wind noise.

On the other hand, when the correlation value K [p] is larger than the predetermined threshold value K _THA , the above-described replacement is not performed, and the L signal and the R signal that have passed through the BPF 30 are output from the signal reduction unit 13 as they are. It is good to give to the signal composition part 15 as a signal and R signal. Alternatively, signal reduction processing as in principle using the multiplier 31 may be performed. That is, when the correlation value K [p] is larger than the predetermined threshold value K _THA , the multiplier 31 is used to reduce the level of the L signal and the R signal that have passed through the BPF 30 to a reduction rate according to the correlation value K [p]. The signal obtained by this may be used as the output signal of the signal reduction unit 13.

<< Second Example >>
Next, a second embodiment will be described. In the second embodiment, both signal restoration processing and signal reduction processing are executed on the frequency axis.

  FIG. 5 is an internal block diagram of the wind noise reduction unit 6b according to the second embodiment. The wind noise reduction unit 6b is used as the wind noise reduction unit 6 of FIG. The wind noise reduction unit 6b includes a correlation value calculation unit 51 that functions as a wind noise determination unit for low band, a wind noise determination unit 52 for medium band, a signal reduction unit 53, a signal restoration unit 54, and a signal synthesis. Part 55. The wind noise determination unit 52 includes n correlation value calculation units 52_1, 52_2,..., 52_n, and the signal reduction unit 53 includes n multipliers 53_1, 53_2,. Is an integer of 2 or more). The signal restoration unit 54 includes each part referred to by the restoration signal generation unit 61 and the signal selector 62.

  The input signal (input acoustic signal) to the wind noise reduction unit 6b is an acoustic signal on the frequency axis composed of a plurality of channel signals (in other words, an acoustic signal expressed in the frequency domain). The input signal to the wind noise reduction unit 6b is an acoustic signal on the frequency axis by time-frequency conversion of the input signals (L (t) and R (t)) to the wind noise reduction unit 6a in FIG. It is converted into a signal. For this time-frequency conversion, DFT (Discrete Fourier Transform), DCT (Discrete Cosine Transform), or the like is used.

  By the time-frequency conversion, the L signal L (t) and the R signal R (t) sampled in the time axis direction every time Δt, and the L signal L (f) sampled in the frequency axis direction every frequency Δf and Converted to R signal R (f). Channel signals corresponding to L (t) and L (f) are called L signals, and channel signals corresponding to R (t) and R (f) are called R signals.

  The input signal to the wind noise reduction unit 6b in FIG. 5 is formed from the L signal L (f) and the R signal R (f) as described above. This input signal is corrected by the wind noise reduction unit 6b. Therefore, an input signal to the wind noise reduction unit 6b is referred to as an “original signal”, and an output signal of the wind noise reduction unit 6b is referred to as a “correction signal”. Hereinafter, the L signal L (f) and the R signal R (f) may be abbreviated as signals L (f) and R (f), respectively.

  Now, for the sake of concrete explanation, it is assumed that a modified discrete cosine transform (MDCT) is used as the time-frequency transform. When MDCT is used, each channel signal on the time axis is divided into frames that are processing units of encoding. One frame includes one or more blocks. Now, it is assumed that one frame is formed from one block. The frame number (that is, the block number) is represented by m, and the m-th frame from 0 is represented as the m-th frame. m takes an integer value of 0 or more. FIG. 6 shows the relationship between the frames. Time advances in the order of the 0th frame, the first frame, the second frame,. Each block has an overlap portion that is half the length of the previous block. In the case of the present example, since one frame is formed from one block, each frame also has an overlap portion that is half the length of one frame with the immediately preceding frame.

  An N sample signal L (t) is converted into an M sample signal L (f) and an N sample signal R (t) is converted into an M sample signal R (f). Assume that 2048 and M = 1024. Further, assuming that the sampling frequency is 48 kHz, the above-described Δt is the reciprocal of 48 kHz. Further, k is introduced as a variable representing the frequency number. Since M = 1024, k takes an integer value of 0 or more and 1023 or less, and Δt = 1/48 kHz, so that frequency interval of the frequency spectrum represented by L (f) and R (f), that is, frequency The frequency interval between the numbers (k-1) and k is about 23 Hz. Therefore, 300 Hz considered as the upper limit of the low band corresponds to k = 13, and 1.5 kHz considered as the upper limit of the medium band corresponds to k = 64.

Then, the signals L (f) and R (f) can be expressed by MDCT coefficients L _{m, k} and R _{m, k} . The MDCT coefficient L _{m, k} represents the signal strength of the frequency component of the frequency number k in the mth frame of the signal L (f), and the MDCT coefficient R _{m, k} represents the _mth of the signal R (f). It represents the signal strength of the frequency component of frequency number k in the frame.

Of the signals L (f) and R (f) forming the original signal, a signal whose frequency band belongs to a low band is input to the correlation value calculation unit 51. That is, the MDCT coefficients L _{m, k} and R _{m, k} within the range of 0 ≦ k ≦ 13 are input to the correlation value calculation unit 51. The correlation value calculation unit 51 calculates a correlation value K _A [m] for each frame according to the following equation (3). K _A [m] represents a correlation value for the m-th frame. K _A [m] is a value between 0 and 1. Needless to say, when a signal is handled on the frequency axis as in the present embodiment, the signal exists at a constant frequency interval, and thus the LPF or the like necessary in the first embodiment is unnecessary.

Wind noise has no cross-correlation between left and right channel signals. Therefore, the correlation value is relatively small if the wind noise contained in the original signal is relatively large, and the correlation value is relatively large if the wind noise contained in the original signal is relatively small. The correlation value K _A [m] takes a value corresponding to the strength of wind noise in the m-th frame. Using this, the correlation value calculation unit 51 as a low-band wind noise determination unit determines the degree of influence of wind noise on each frame based on the correlation value. This determination result is used for processing in the signal restoration unit 54.

Of the signals L (f) and R (f) forming the original signal, a signal whose frequency band belongs to the middle band is input to the wind noise determination unit 52 and the signal reduction unit 53. That is, MDCT coefficients L _{m, k} and R _{m, k} within the range of 14 ≦ k ≦ 64 are input to the wind noise determination unit 52 and the signal reduction unit 53. The input signals to the wind noise determination unit 52 and the signal reduction unit 53 are further divided into n pieces. That is, the middle band is further divided into n pieces, and wind noise determination and signal reduction are performed for each subdivided band.

In particular,
MDCT coefficients L _{m, k} and R _{m, k} within the range of 14 ≦ k ≦ k ₁ are input to the correlation value calculation unit 52_1 and the multiplier 53_1,
MDCT coefficients L _{m, k} and R _{m, k} within the range of k ₁ <k ≦ k ₂ are input to the correlation value calculator 52_2 and the multiplier 53_2,...
k _n over 1 <MDCT coefficient in the range of _k ≦ k _n L _m, k and R _{m, k} are input to the correlation value calculation section 52_n and the multiplier 53_N.
Here, 14 <k ₁ <k ₂ <,... <K _n-1 <k _n = 64.

The wind noise determination unit 52 calculates a correlation value for each of the n subdivided bands. That is, the correlation value calculation unit 52_1 calculates the correlation value K _B1 [m] for each frame according to the following equation (4-1), and the correlation value calculation unit 52_2 for each frame according to the following equation (4-2). The value K _B2 [m] is calculated, and the correlation value calculation unit 52_n calculates the correlation value K _Bn [m] for each frame according to the following equation (4-n). Correlation values K _B1 [m], K _B2 [m],... K _Bn [m] represent correlation values for the m-th frame. K _B1 [m], K _B2 [m],..., K _Bn [m] represent the cross-correlation between the L signal and the R signal for the corresponding band, and are values of 0 or more and 1 or less, respectively. .

For the m-th frame, the multiplier 53_1 uses the level of the input signal to itself (that is, the value of the MDCT coefficients L _{m, k} and R _{m, k} within the range of 14 ≦ k ≦ k ₁ ) as the correlation value K _B1 [ m] and output a signal after the reduction. Similarly,
For the m-th frame, the multiplier 53_2 uses the level of the input signal to itself (that is, the values of the MDCT coefficients L _{m, k} and R _{m, k} within the range of k ₁ <k ≦ k ₂ ) as the correlation value K _B2. The signal is reduced at a reduction rate according to [m], and the signal after reduction is output. Similarly,
Multiplier 53_n relates frame of the m, the level of the input signal to itself (i.e., k _n over 1 <MDCT coefficient in the range of _k ≦ k _n L _m, k and R _m, the value of _k) correlation values The signal is reduced at a reduction rate corresponding to K _Bn [m], and the signal after reduction is output.
The same applies to the other multipliers in the signal reduction unit 53.

At this time, when j is an integer of 1 to n, the multiplier 53_j greatly reduces the level when the influence of wind noise is determined to be large from the correlation value K _Bj [m], and the correlation value K _Bj When it is determined from [m] that the influence of wind noise is small, the level is not reduced so much. That is, the multiplier 53_j increases the reduction rate corresponding to the m-th frame as the correlation value K _Bj [m] decreases, and decreases corresponding to the m-th frame as the correlation value K _Bj [m] increases. Decrease rate. The greater the reduction rate, the greater the level reduction level in the multiplier 53_j. The level to be reduced is specifically the values of the MDCT coefficients L _{m, k} and R _{m, k} within the range of 14 ≦ k ≦ k ₁ when, for example, j = 1.

In addition, if the same result is led, the method of reducing is arbitrary. For example, the correlation value calculated by the correlation value calculation unit 52_j or a coefficient corresponding to the correlation value may be multiplied by the input signal of the multiplier 53_j. When the correlation value K _Bj [m] is larger than a predetermined reference threshold, it may be determined that there is no wind noise and the input signal of the multiplier 53_j may be used as the output signal of the multiplier 53_j as it is.

  In the middle band, the band affected by wind noise differs depending on the wind intensity and the like. Therefore, the influence level of wind noise is evaluated by correlation value calculation for each subdivided band by subdividing the middle band into smaller bands. Then, for each subdivided band, the signal level reduction degree is adjusted according to the wind noise influence degree. As a result, signal reduction is performed only for the subdivided band that is affected by wind noise, or stronger signal reduction is performed for the subdivided band that is more affected by wind noise. As a result, a wind noise reduction effect for the middle band can be obtained without unnecessary signal reduction.

  The output signals of the multipliers 53_1, 53_2,..., 53_n are combined, and the MDCT coefficient in the middle band obtained by the combination is generated as an output signal (second corrected acoustic signal) of the signal reduction unit 53. Sent to the unit 61 and the signal synthesis unit 55.

  The restoration signal generation unit 61 restores a low-band acoustic signal on the frequency axis by predicting a low-band pitch from the middle-band pitch information included in the output signal of the signal reduction unit 53. This restoration method will be described by focusing on one frame. Please refer to FIG. A solid broken line denoted by reference numeral 300 in FIG. 7 represents the frequency spectrum of the middle band in the frame of interest given to the restoration signal generation unit 61. In the present embodiment, the frequency spectrum 300 is defined by the output signal of the signal reduction unit 53.

In FIG. 7, the horizontal axis represents the frequency, and the vertical axis represents the level of the frequency spectrum. The level of the frequency spectrum is represented by the value of the MDCT coefficient. FIG. 7 corresponds to the case where a pitch is included in the frame of interest. If a pitch is included in the frame of interest, the frequency spectrum periodically changes, and takes a local maximum value and a local minimum value periodically. Now, it is assumed that the frequency spectrum 300 has maximum values at frequencies f _A , f _C , f _E and f _G and has minimum values at frequencies f _B , f _D , f _F and f _H. Here, it is assumed that f _A <f _B <f _C <f _D <f _E <f _F <f _G <f _H.

The restoration signal generation unit 61 detects the frequencies f _A , f _B , f _C , f _D , f _E , f _F , f _G, and f _H from the frequency spectrum 300, and the difference value between the adjacent maximum value and the minimum value. When the difference value is equal to or greater than a predetermined difference threshold value, it is determined that the frequency component having the maximum value corresponding to the difference value is a harmonic component of the pitch. For example, the difference value obtained by subtracting the level of the frequency f _B from the level of the frequency f _A in the frequency spectrum 300 is compared with the above difference threshold, and if the former is greater than or equal to the latter, the component of the frequency f _A While it is determined that the component is a harmonic component, when the former is less than the latter, it is determined that the component of the frequency f _A is not a harmonic component of the pitch. The same applies to the frequencies corresponding to the other maximum and minimum values.

Now, it is assumed that the frequencies f _A , f _C , f _E and f _G are determined to be harmonic components of the pitch by this determination. In this case, the restoration signal generation unit 61 predicts the pitch interval Dp from the frequency difference between adjacent harmonic components. For example, an average of the frequency differences (f _A −f _C ), (f _C −f _E ), and (f _E −f _G ) is set as the pitch interval Dp. Further, the restoration signal generation unit 61 predicts the pitch level Gp from the levels of the frequencies f _A , f _C , f _E and f _{G in} the frequency spectrum 300.

The restoration signal generation unit 61 predicts a low-band signal from the pitch information including the predicted pitch interval Dp and pitch level Gp, and generates a restoration signal. That is, it is predicted that there is a pitch at a frequency f _X (= f _A −Dp) that is lower than the frequency of the harmonic component of the pitch on the lowest frequency side in the middle band by a pitch interval Dp, and the level is at that frequency f _X. Restore the pitch of Gp. The state of this restoration is shown in FIGS. In FIG. 7, a broken line with a reference numeral 301 represents a frequency spectrum of a low-band restoration signal on the frequency axis generated by the restoration signal generation unit 61.

The level Gp is calculated by linearly interpolating or curve interpolating the levels of the frequencies f _A , f _C , f _E and f _{G in} the frequency spectrum 300 on the coordinate plane representing the frequency spectrum 300. For example, if the levels of frequencies f _A , f _C , f _E and f _G are quantified by 10, 8, 6 and 4, respectively, Gp is predicted to be 12.

In addition, a restoration signal other than the frequency f _X (that is, the polygonal line shape of the frequency spectrum 301 in FIG. 7) is predicted so that the level gradually decreases as the frequency increases from the frequency f _X. In this prediction, the frequency spectrum 300 may be taken into consideration. For example, a restored signal other than the frequency f _X may be predicted in consideration of the spectrum shape between adjacent maximum and minimum values in the frequency spectrum 300. For example, the spectrum shape between the frequencies f _B -f _D in the frequency spectrum 300 in the level direction at the ratio between the level Gp and the level at the frequency f _C (12/8 = 1.5 in the above numerical example). The spectrum shape obtained by stretching may be the shape of the frequency spectrum 301. In the example shown in FIGS. 7 and 8, there is only one restored pitch, but when the calculated pitch interval Dp is narrow, a restored signal is generated so that a plurality of pitches exist in the low band. It doesn't matter if you do.

The signal selector 62 is given a low-band signal in the original signal and the restored signal generated by the restored signal generator 61, and the signal selector 62 selects one of the former and the latter for each frame. _A correlation value K _A [m] calculated by the correlation value calculation unit 51 is selected and output. Both the low-band signal in the original signal and the restored signal generated by the restored signal generator 61 are expressed by MDCT coefficients L _{m, k} and R _{m, k} within the range of 0 ≦ k ≦ 13. The values of the MDCT coefficients L _{m, k} and R _{m, k} between are usually different.

Specifically, when focusing on the m-th frame, the signal selector 62 compares the correlation value K _A [m] with a predetermined threshold, and when the correlation value K _A [m] is equal to or smaller than the predetermined threshold. It is determined that there is wind noise, and a restoration signal corresponding to the mth frame is selected and output. If the correlation value K _A [m] is larger than the threshold, it is determined that there is no wind noise and the mth frame. A low-band signal in the original signal corresponding to is selected and output. The output signal of the signal selector 62 is used as the output signal (first corrected acoustic signal) of the signal restoration unit 54.

  In addition to the output signals of the signal restoration unit 54 and the signal reduction unit 53, the signal synthesis unit 55 is given the high-band signal in the original signal as it is. The signal synthesis unit 55 outputs, for each frame, an output signal of the signal restoration unit 54 that represents a low-band acoustic signal and an output signal of the signal reduction unit 53 that represents a mid-band acoustic signal in which wind noise has been reduced by the signal reduction processing. Are combined with a high-band signal in the original signal, and a signal obtained by the synthesis is output as an output signal (that is, a correction signal) of the wind noise reduction unit 6b. In the second embodiment, this correction signal is also an acoustic signal on the frequency axis composed of a plurality of channel signals, like the original signal.

  In the audio signal processing unit 4 of FIG. 2, the correction signal output from the signal synthesis unit 55 is quantized according to the AAC encoding method and converted into a bit stream as an encoded audio signal. This encoded audio signal (bit stream) is recorded on the recording medium 5 of FIG.

  In the above description, in principle, the signal processing for the L signal and the R signal is not described separately. However, as described above, each part in the wind noise reduction unit 6b is assigned to each of a plurality of channel signals. On the other hand, necessary signal processing is performed individually.

  That is, the multiplier 53_j performs signal reduction processing corresponding to the correlation value calculated by the correlation value calculation unit 52_j for each of the L signal and the R signal in the middle band in the original signal (as described above, j is An integer of 1 to n). The restoration signal generation unit 61 creates pitch information of each of the L signal and the R signal that form the output signal of the signal reduction unit 53, and generates a restoration signal of the L signal and the R signal based on each pitch information. The signal selector 62 selects and outputs the L signal and R signal in the low band in the original signal or the L signal and R signal in the restored signal according to the correlation value calculated by the correlation value calculation unit 51. The signal synthesis unit 55 synthesizes the L signal in the output signal of the signal restoration unit 54, the L signal in the output signal of the signal reduction unit 53, and the L signal in the high band in the original signal, and in the output signal of the signal restoration unit 54 The correction signal is generated by combining the R signal and the R signal in the output signal of the signal reduction unit 53 and the R signal in the high band in the original signal.

  In this embodiment, the restoration signal is generated based on the output signal of the signal reduction unit 53 (that is, the signal in the middle band after the signal reduction process), but the signal in the middle band in the original signal is generated. It is also possible to generate a restoration signal based on this. In this case, instead of the output signal of the signal reduction unit 53, a medium band signal in the original signal may be supplied to the restoration signal generation unit 61. However, if pitch information is extracted after reducing wind noise by signal reduction processing, more accurate information can be obtained. Therefore, it is desirable to configure as shown in FIG.

  Further, the correlation value calculation unit 51 can be omitted from the wind noise reduction unit 6b. When the correlation value calculation unit 51 is omitted, the signal selector 62 is also omitted, and the signal restoration unit 54 outputs the restoration signal generated by the restoration signal generation unit 61 unconditionally. In a similar manner, the wind noise determination unit 52 can be omitted from the wind noise reduction unit 6b. When the wind noise determination unit 52 is omitted, the multiplier 53_j reduces the signal level in the middle band in the original signal at a constant reduction rate, and outputs the signal after the reduction. If the correlation value calculation unit 51 and the wind noise determination unit 52 are omitted, the input signal to the wind noise reduction unit 6b may be a monaural signal composed of one channel signal.

Further, the wind noise reduction unit 6 b is provided with a low-band wind noise determination unit and a mid-band wind noise determination unit 52 as the correlation value calculation unit 51 independently, and the former determination result is obtained from the signal restoration unit 52. Although only the processing is reflected and the latter determination result is reflected only on the processing of the signal reduction unit 53, each determination result may be mutually used as follows. That is, for example, based on the correlation value K _A [m] calculated by the correlation value calculation unit 51 and the correlation value K _Bj [m] calculated by the correlation value calculation unit 52_j, the reduction rate in the multiplier 53_j in the m-th frame May be determined. More specifically, for example, not only the reduction rate is increased according to the decrease of the correlation value K _Bj [m], but also the reduction rate is increased according to the decrease of the correlation value K _A [m]. Similarly, the selection of the signal selector 62 in the m-th frame is performed based on the correlation value K _A [m] calculated by the correlation value calculation unit 51 and the correlation value K _Bj [m] calculated by the correlation value calculation unit 52_j. You may make it do.

[Modification of signal reduction processing]
In the signal reduction process described above, the signal levels of the L signal and the R signal in the middle band of the original signal are reduced using the multiplier 53_j, and the signal after this reduction is given to the signal synthesis unit 55. However, instead of this, the following processing may be executed. As an example, third and fourth modified signal reduction processes related to the signal reduction process of the second embodiment will be exemplified. The third and fourth modified signal reduction processes correspond to processes in which the first and second modified signal reduction processes described in the first embodiment are adapted to the second embodiment, respectively.

The third deformation signal reduction process will be described. For the sake of specific description, first, attention is focused on a band corresponding to the correlation value calculation unit 52_1, which is one of the n subdivided bands. In the third modified signal reduction process, the signal reduction unit 53 compares the correlation value K _B1 [m] calculated by the correlation value calculation unit 52_1 with a predetermined threshold value K _THB1 . As described above, the correlation value K _B1 [m] represents the degree of influence of wind noise on the specific band in the m-th frame. On the other hand, the threshold value K _THB1 represents a reference influence degree to be compared with the influence degree. When the correlation value K _B1 [m] is smaller than the threshold value K _THB1 , it is determined that the influence of wind noise on the specific band in the m-th frame is relatively large, and the correlation value K _B1 [m] is smaller than the threshold value K _THB1. If it is larger, it is determined that it is relatively small (the same applies to the correlation values K _B2 [m] to K _Bn [m] and the same applies to the fourth modified signal reduction process).

When the correlation value K _B1 [m] is smaller than the threshold value K _THB1 , the signal reduction unit 53 determines the MDCT coefficients L _{m, k} and R _{m, k} within the range of 14 ≦ k ≦ k ₁ included in the original signal. MDCT coefficient (L _{m, k} + R _{m, k} ) / 2 is calculated, and the MDCT coefficient (L _{m, k} + R _{m, k} ) / 2 is output from the signal reduction unit 53 to 14 ≦ Treated as MDCT coefficients L _{m, k} and R _{m, k} in the range of k ≦ k ₁ .

On the other hand, when the correlation value K _B1 [m] is larger than the threshold value K _THB1 , the above averaging is not performed, and the MDCT coefficients L _{m, k} and R within the range of 14 ≦ k ≦ k ₁ included in the original signal. _{m, k} is handled as MDCT coefficients L _{m, k} and R _{m, k} within the range of 14 ≦ k ≦ k ₁ to be output from the signal reduction unit 53 as they are (or the above signal reduction processing by the multiplier 53_1 is performed). You may go).

The above processing is performed individually for each of the subdivided n bands. Generalize using variable j. The MDCT coefficients L _{m, k} and R _{m, k} within the range of k _j−1 ≦ k ≦ k _j to be output from the signal reduction unit 53 are respectively converted into the MDCT coefficients L _{m, k} ′ and R _{m, k} ′. Represented by As described above, it is assumed that 14 <k ₁ <k ₂ <,... <K _n-1 <k _n = 64, and k ₀ = 14.

The signal reduction unit 53 compares the correlation value K _Bj [m] calculated by the correlation value calculation unit 52 — j with a predetermined threshold value K _THBj for each of j = 1, 2 _,. When the correlation value K _Bj [m] is smaller than the threshold value K _THBj , the signal reduction unit 53 includes the MDCT coefficients L _{m, k} and R within the range of k _j−1 ≦ k ≦ k _j included in the original signal. _m, averaged MDCT coefficient _{k (L m, k + R} m, k) / 2 is calculated, the MDCT coefficients _{(L m, k + R m} , k) / 2 _{a, k j-1 ≦ k ≦} k Output as MDCT coefficients L _{m, k} ′ and R _{m, k} ′ within the range of _j . On the other hand, when the correlation value K _Bj [m] is larger than the threshold value K _THBj , the above averaging is not performed, and the MDCT coefficient L _m, within the range of k _j−1 ≦ k ≦ k _j included in the original signal _{. k} and R _{m, k} are output as they are as MDCT coefficients L _{m, k} ′ and R _{m, k} ′ within the range of k _j−1 ≦ k ≦ k _j (or the above signal reduction processing by the multiplier 53_j is performed). You may go).

  If the above averaging is performed, the influence of wind noise is equalized between the channels, and the noise level of a channel that is relatively affected by wind noise is reduced. Further, by performing signal reduction processing for each subdivided band, it is possible to efficiently reduce the noise level only in the band affected by wind noise.

A fourth modified signal reduction process will be described. For the sake of specific description, first, attention is focused on a band corresponding to the correlation value calculation unit 52_1, which is one of the n subdivided bands. In the fourth modified signal reduction process, the signal reduction unit 53 compares the correlation value K _B1 [m] calculated by the correlation value calculation unit 52_1 with a predetermined threshold value K _THB1 . When the correlation value K _B1 [m] is smaller than the threshold value K _THB1 , the signal level among the MDCT coefficients L _{m, k} and R _{m, k} within the range of 14 ≦ k ≦ k ₁ included in the original signal. Is specified as the minimum acoustic signal and the other signal (that is, the MDCT coefficient with the larger absolute value) is specified as the non-minimum acoustic signal. Replace the signal with the minimum acoustic signal.

That is, when the correlation value K _B1 [m] is smaller than the threshold value K _THB1 , the MDCT coefficients L _{m, k} and R _{m, k} within the range of 14 ≦ k ≦ k ₁ included in the original signal, for example, MDCT When the coefficient R _{m, k} is specified as the minimum acoustic signal, the MDCT coefficient R _{m, k} representing the minimum acoustic signal is the MDCT coefficient L _{m, k} ′ within the range of 14 ≦ k ≦ k ₁ , and , 14 ≦ k ≦ k ₁ , and output as MDCT coefficients R _{m, k} ′.

On the other hand, when the correlation value K _B1 [m] is larger than the threshold value K _THB1 , the above replacement is not performed, and MDCT coefficients L _{m, k} and R _m within the range of 14 ≦ k ≦ k ₁ included in the original signal are not performed. _{, k} are output as MDCT coefficients L _{m, k} ′ and R _{m, k} ′ within the range of 14 ≦ k ≦ k ₁ (or the above signal reduction processing by the multiplier 53_1 may be performed).

The above processing is performed individually for each of the subdivided n bands. Generalize using variable j. The signal reduction unit 53 compares the correlation value K _Bj [m] calculated by the correlation value calculation unit 52 — j with a predetermined threshold value K _THBj for each of j = 1, 2 _,. When the correlation value K _Bj [m] is smaller than the threshold value K _THBj , the MDCT coefficients L _{m, k} and R _{m, k} within the range of k _j−1 ≦ k ≦ k _j included in the original signal are included. The signal having the smaller signal level (ie, the MDCT coefficient having the smaller absolute value) is specified as the minimum acoustic signal, and the other signal (that is, the MDCT coefficient having the larger absolute value) is specified as the non-minimum acoustic signal. Replace non-minimum acoustic signal with minimum acoustic signal. Then, the replaced MDCT coefficients are output as MDCT coefficients L _{m, k} ′ and R _{m, k} ′ within the range of k _j−1 ≦ k ≦ k _j .

On the other hand, when the correlation value K _Bj [m] is larger than the threshold value K _THBj , the above replacement is not performed, and the MDCT coefficient L _{m, k} within the range of k _j−1 ≦ k ≦ k _j included in the original signal. And R _{m, k} are output as MDCT coefficients L _{m, k} ′ and R _{m, k} ′ within the range of k _j−1 ≦ k ≦ k _j (or the above signal reduction processing by the multiplier 53_j is performed). May be)

  By performing the above replacement, it is possible to reduce the noise level of a channel relatively affected by wind noise without increasing the noise level of a channel relatively affected by wind noise. Further, by performing signal reduction processing for each subdivided band, it is possible to efficiently reduce the noise level only in the band affected by wind noise.

MDCT coefficients L _{m, k} ′ and R _{m, k} ′ within the range of 14 ≦ k ≦ 64 obtained by performing the third or fourth modified signal reduction processing are synthesized, and the intermediate band obtained by the synthesis These MDCT coefficients are sent to the restoration signal generation unit 61 and the signal synthesis unit 55 as output signals of the signal reduction unit 53.

Further, the threshold value K _THBj may be variably set based on the calculation result of the correlation value calculation unit 51 that functions as a low-band wind noise determination unit. Specifically, for example, the threshold value K _THBj is variably set so that the above averaging or replacement is easily performed as the correlation value K _A [m] obtained by the correlation value calculation unit 51 becomes smaller. That is, as the correlation value K _A [m] decreases, the threshold value K _THBj to be compared with the correlation value K _Bj [m] is increased.

<< Third Example >>
Next, a third embodiment will be described. In the third embodiment, the signal restoration process is executed on the time axis, and then the signal reduction process is executed on the frequency axis after performing time-frequency conversion. By performing each process in an easily realizable area (time domain or frequency domain), it is possible to form a wind noise reduction unit with higher accuracy and less processing load.

  FIG. 9 is an internal block diagram of the wind noise reduction unit 6c according to the third embodiment. The wind noise reduction unit 6c is used as the wind noise reduction unit 6 of FIG. The wind noise reduction unit 6c includes a wind noise determination unit 11 that functions as a wind noise determination unit for low band, a signal restoration unit 12, a wind noise determination unit 52 that functions as a wind noise determination unit for medium band, and a signal A reduction unit 53, an HPF 81, a signal synthesis unit 82, a time frequency conversion unit 83, and a signal synthesis unit 84 are provided.

  The input signals (input acoustic signals) to the wind noise reduction unit 6c are the same signals L (t) and R (t) as the input signals to the wind noise reduction unit 6a in FIG. This input signal is corrected by the wind noise reduction unit 6c. Accordingly, an input signal to the wind noise reduction unit 6c is referred to as an “original signal”, and an output signal of the wind noise reduction unit 6c is referred to as a “correction signal”.

  In the third embodiment, the original signal is input to each of the BPF 23, the LPFs 21 and 26, and the HPF 81.

  The wind noise determination unit 11 and the signal restoration unit 12 in the wind noise reduction unit 6c are the same as those in the wind noise reduction unit 6a in FIG. That is, according to the correlation value calculated by the wind noise determination unit 11, the signal restoration unit 12 performs weighted addition of the low-band signal of the original signal and the low-band signal of the restoration signal, and thereby the output of the signal restoration unit 12 A signal (first corrected acoustic signal) is generated.

  The HPF 81 passes only the middle band component and the high band component of the input signal to itself.

  The signal synthesis unit 82 adds the output signal of the signal restoration unit 12 that represents a low-band acoustic signal in which wind noise has been reduced by the signal restoration process, and the output signal of the HPF 81, and the signal obtained by this addition is It outputs to the time frequency conversion part 83. In addition, when each signal delay amount in the signal restoration unit 12 and the HPF 81 is different, after performing delay processing to cancel the difference between the signal delay amounts in the signal synthesis unit 82 or in the previous stage of the signal synthesis unit 82, The addition processing of the signal synthesis unit 82 is performed. The same applies to the weighted addition processing using the multipliers 27 and 28 and the adder 29.

  The acoustic signal output from the signal synthesis unit 82 is an acoustic signal on the time axis composed of an L signal and an R signal. Each value of the L signal and R signal forming the output signal of the signal synthesis unit 82 is different from that of the L signal and R signal forming the original signal, but the L signal and the R signal forming the output signal of the signal synthesis unit 82 are different. The R signal is also expressed by L (t) and R (t) for convenience of explanation.

  The time frequency conversion unit 83 converts the output signal of the signal synthesis unit 82 into a signal on the frequency axis by time frequency conversion. This time-frequency conversion is the same as the time-frequency conversion described in the second embodiment. That is, the time-frequency conversion unit 83 forms the output signal of the signal synthesis unit 82 by time-frequency conversion, and the L signal L (t) and the R signal R (t) sampled in the time axis direction every time Δt. The L signal L (f) and the R signal R (f) sampled in the frequency axis direction for each frequency Δf are converted and output. Since the signal restoration processing for the low band is performed in the previous stage of the time frequency conversion unit 83, the values of the low band components of the L signal L (f) and the R signal R (f) obtained by this conversion are shown in FIG. This is different from that of the original signal (L (f) and R (f) in the second embodiment) for the wind noise reduction unit 6b, but is output from the time-frequency conversion unit 83 in this embodiment for convenience of explanation. The L signal and the R signal are expressed as L (f) and R (f).

For the sake of concrete explanation, it is assumed that the modified discrete cosine transform (MDCT) is used as the time-frequency transform in the time-frequency transform unit 83 as in the second embodiment. The specific example for MDCT described in the second example is also applied to this example (specific examples of values such as N, M, m, and k are also applied). Then, the signals L (f) and R (f) forming the output signal of the time frequency conversion unit 83 can be expressed by MDCT coefficients L _{m, k} and R _{m, k} .

Of the signals L (f) and R (f) forming the output signal of the time-frequency conversion unit 83, a signal whose frequency band belongs to the middle band is input to the wind noise determination unit 52 and the signal reduction unit 53. That is, MDCT coefficients L _{m, k} and R _{m, k} within the range of 14 ≦ k ≦ 64 are input to the wind noise determination unit 52 and the signal reduction unit 53.

  The wind noise determination unit 52 and the signal reduction unit 53 in the wind noise reduction unit 6c are the same as those in the wind noise reduction unit 6b in FIG. That is, the middle band is subdivided into n, and the middle band of the output signal of the time-frequency conversion 83 is reduced for each subdivided band at a reduction rate according to the correlation value calculated by the wind noise determination unit 52. Let The output signals of the multipliers 53_1, 53_2,..., 53_n, which are the signals after the reduction, are combined, and the midband MDCT coefficient obtained by the combination is the output signal of the signal reduction unit 53 (second signal). Corrected acoustic signal) is sent to the signal synthesizer 84. It should be noted that the output signals of the multipliers 53_1, 53_2,..., 53_n may be combined by the signal combining unit 84.

  Of the signals L (f) and R (f) forming the output signal of the time-frequency conversion unit 83, signals belonging to the low and high frequency bands are supplied to the signal synthesis unit 84 as they are. The signal synthesizer 84 synthesizes the signal belonging to the low band and the high band directly given from the time-frequency converter 83 and the output signal of the signal reducer 53 for each frame, and the signal obtained by this synthesis. Is output as an output signal (that is, a correction signal) of the wind noise reduction unit 6c. In the third embodiment, the correction signal is an acoustic signal on the frequency axis composed of a plurality of channel signals.

  In the audio signal processing unit 4 of FIG. 2, the correction signal output from the signal synthesis unit 84 is quantized according to the AAC encoding method and converted into a bit stream as an encoded audio signal. This encoded audio signal (bit stream) is recorded on the recording medium 5 of FIG.

  In the above description, in principle, the signal processing for the L signal and the R signal is not separately described. However, as described above, each part in the wind noise reduction unit 6c is assigned to each of a plurality of channel signals. On the other hand, necessary signal processing is performed individually.

  That is, the HPF 81 passes only the middle band and high band components of the L signal forming the original signal and the middle band and high band components of the R signal forming the original signal. The signal synthesis unit 82 adds the L signal in the output signal of the signal restoration unit 12 and the L signal in the output signal of the HPF 81, and adds the R signal in the output signal of the signal restoration unit 12 and the R signal in the output signal of the HPF 81. To do. The time-frequency conversion unit 83 performs time-frequency conversion on the given L and R signals on the time axis individually. The signal synthesis unit 84 synthesizes the L signal in the output signal of the signal reduction unit 53 and the L signal in the low band and the high band in the output signal of the time frequency conversion unit 83 and also the R signal in the output signal of the signal reduction unit 53. And the R signal in the low band and the high band in the output signal of the time-frequency conversion unit 83 are combined to generate a correction signal. The LPF 21 and the like are as described in the first or second embodiment.

  Further, as described in the first embodiment, the wind noise determination unit 11 can be omitted from the wind noise reduction unit 6c. When the wind noise determination unit 11 is omitted, the multipliers 27 and 28 and the adder 29 perform weighted addition of the output signal values of the LPFs 25 and 26 at a constant ratio, thereby generating an output signal (first signal) of the signal restoration unit 12. Corrected acoustic signal) is generated. Further, as described in the second embodiment, the wind noise determination unit 52 can be omitted from the wind noise reduction unit 6c. When the wind noise determination unit 52 is omitted, the multiplier 53_j reduces the signal level in the middle band in the output signal of the time-frequency conversion unit 83 at a constant reduction rate, and outputs the signal after the reduction. When the wind noise determination unit 11 and the wind noise determination unit 52 are omitted, the input signal to the wind noise reduction unit 6c may be a monaural signal composed of one channel signal.

  Further, the wind noise reduction unit 6c is provided with the wind noise determination unit 11 for low band and the wind noise determination unit 52 for medium band independently, and the former determination result is reflected only in the processing of the signal restoration unit 12; Although the latter determination result is reflected only in the processing of the signal reduction unit 53, each determination result may be used mutually as described in the second embodiment. That is, for example, for a certain frame of interest, the reduction rate in the multiplier 53_j in the frame of interest is determined based on the correlation value calculated by the correlation value calculation unit 22 and the correlation value calculated by the correlation value calculation unit 52_j. More specifically, for example, not only the reduction rate is increased according to the decrease of the correlation value calculated by the correlation value calculation unit 52_j, but also the reduction rate is increased according to the decrease of the correlation value calculated by the correlation value calculation unit 22. Like that.

Further, the third and fourth modified signal reduction processes described in the second embodiment can be applied to the third embodiment. Needless to say, when the third and fourth modified signal reduction processes are applied to the third embodiment, the “original signal and signal combining unit 55” described in the description of the third and fourth modified signal reduction processes is provided. Are read as “the output signal of the time-frequency converter 83 and the signal synthesizer 84. Also, when the third and fourth modified signal reduction processes are applied to the third embodiment, the wind noise for the low band is used. The threshold value K _THBj may be variably set according to the determination result of the determination unit 11. Specifically, for example, for a certain frame of interest, as the correlation value obtained by the correlation value calculation unit 22 decreases. The threshold value K _THBj is variably set so that the above averaging or replacement is easily performed, that is, as the correlation value obtained by the correlation value calculation unit 22 for a certain frame of _interest decreases, the correlation value K is compared with _Bj [m] A larger threshold K _THBj should.

  According to the first to third embodiments, it is possible to eliminate the distortion of the low-band signal caused by the wind noise reduction process. In addition, the influence of wind noise can be suppressed also in the middle band by the signal reduction process.

Here, the advantages of the respective wind noise reduction units of the first to third embodiments will be considered.
In the wind noise reduction unit 6a (FIG. 3) according to the first embodiment, signal restoration processing and signal reduction processing can be performed in parallel. In addition, since signal processing can be performed only in the time domain, time-frequency conversion is not necessary.
Since the wind noise reduction unit 6b (FIG. 5) according to the second embodiment performs signal processing in the frequency domain, it can intuitively perform processing for each band. It is possible to easily divide the middle band on which signal reduction processing is performed, and it is possible to perform signal reduction only in a band affected by wind.
The wind noise reduction unit 6c (FIG. 9) according to the third embodiment is easy to incorporate into an encoder (encoder) according to AAC or the like, and has high practicality.

<< 4th Example >>
As described above, the wind noise reduction unit 6c according to the third embodiment can be easily incorporated into an encoder according to AAC or the like. For example, MDCT can be used for time-frequency conversion, or the output correction signal on the frequency axis can be used as it is for quantization processing of the encoder. A fourth embodiment will be described as an embodiment relating to incorporation into an encoder.

  FIG. 10 shows an internal block diagram of the AAC encoder 110 that can be used in combination with the wind noise reduction unit 6c of FIG. The AAC encoder 110 is built in the audio signal processing unit 4 of FIG. Since the operation of each part in the AAC encoder 110 conforms to the AAC standard, description thereof is omitted. The filter bank 111 provided in the AAC encoder 110 is a part that performs the modified discrete cosine transform, and corresponds to the time-frequency transform unit 83 in FIG.

  The wind noise determination unit 11, the signal restoration unit 12, the HPF 81, and the signal synthesis unit 82 in the wind noise reduction unit 6 c of FIG. 9 are provided in the previous stage of the AAC encoder 110, and the output signal of the signal synthesis unit 82 is input to the input signal of the AAC encoder 110. Give as. Then, the signal reduction unit 53 corrects the intermediate band of the output signal of the filter bank 111 corresponding to the output signal of the time-frequency conversion unit 83, and outputs a signal (that is, output from the signal synthesis unit 84 of FIG. 9). Correction signal) is supplied to a portion (TNS (Temporal Noise Shaping) and bit stream multiplexer) that requires the output signal of the filter bank 111. The bit stream output from the AAC encoder 110 via this correction is recorded on the recording medium 5 in FIG.

Note that when the wind noise reduction unit (6b or 6c) is incorporated into an encoder such as the AAC encoder 110, the band division method may be matched to the audio format of the incorporated encoder. Thereby, processing can be simplified. That is, for example, the MDCT coefficients L _{m, k} and R _{m, k} described in the second or third embodiment may be combined with the MDCT coefficients used in the encoder.

  In addition, when the wind noise reduction unit (6b or 6c) is incorporated in an encoder such as the AAC encoder 110, acoustic signals on the time axis may overlap between adjacent frames. That is, for example, in the case of the specific example of MDCT described in the second or third embodiment, as shown in FIG. 6, the acoustic signals on the time axis of 1024 samples are overlapped between adjacent frames. In such a case, in the wind noise determination unit 11 and the signal restoration unit 12 of the wind noise reduction unit 6c, the low-band wind noise determination and the mid-band wind noise determination are equivalent as follows. It is desirable to process.

  That is, the correlation value calculation unit 22 of the wind noise determination unit 11 calculates a correlation value according to the above equation (1) for each frame. This is realized by handling the “unit section” introduced in the first embodiment as a “frame” adapted to MDCT. In this case, although different from the situation shown in FIG. 4, adjacent unit sections overlap each other by half of the unit sections. Then, for example, a correlation value for a certain frame is calculated based on the first to 2048th acoustic signals on the time axis, and thereafter, a correlation for the next frame is calculated based on the 1025th to 3072th acoustic signals on the time axis. A value will be calculated. The multipliers 27 and 28 and the adder 29 convert the output signal values of the LPFs 25 and 26 for 1024 samples of the first half (or second half) of the m-th frame to the correlation values calculated by the correlation value calculation unit 22 for the m-th frame. Accordingly, the weighted addition is performed according to the above equation (2), thereby forming the output signal of the signal restoration unit 12.

<< 5th Example >>
In each of the embodiments described above, correction processing (signal restoration processing and signal reduction processing) for reducing wind noise is performed in real time on the output signal of each microphone, and the correction signal obtained thereby is used as the recording medium 5 in FIG. However, the timing for executing the correction process is arbitrary.

  For example, the original signal on the time axis before correction or the original signal on the frequency axis before correction based on the output signals of the microphones MIC1 and MIC2 is once recorded as raw data on the recording medium 5. Of course, signal processing such as compression processing may be appropriately performed during the recording. Then, at the time of audio reproduction or the like, the original signal on the time axis before correction or the original signal on the frequency axis before correction is reproduced from the raw data, and the reproduced original signal is reproduced in the wind noise reduction unit (6a, 6b or 6c). The correction signal may be obtained by giving the correction signal. Then, at the time of audio reproduction, the correction signal may be reproduced and output.

  As is clear from the above description, it is also possible to mount an audio signal processing unit including a wind noise reduction unit (6a, 6b or 6c) on an acoustic signal reproduction device that reproduces an acoustic signal from the raw data. Even in this case, the wind noise reduction unit functions effectively. That is, the present invention can also be applied to an acoustic signal reproduction device. If raw data is recorded at the time of sound collection and correction processing for reducing wind noise is performed on the sound signal playback device side, it is possible to freely switch execution / non-execution of the correction processing during playback. It is.

  Further, although the imaging apparatus 1 is illustrated as an apparatus provided with the audio signal processing unit 4, a similar audio signal processing unit can be provided in another recording apparatus or an apparatus having a recording function. Other recording devices or devices having a recording function include, for example, portable recording devices such as IC recorders and mobile phones having a recording function. These devices are provided with the microphones MIC1 and MIC2, the audio signal processing unit 4, and the recording medium 5 shown in FIG.

<< Deformation, etc. >>
The specific numerical values shown in the above description are merely examples, and as a matter of course, they can be changed to various numerical values. As modifications or annotations of the above-described embodiment, notes 1 to 4 are described below. The contents described in each comment can be arbitrarily combined as long as there is no contradiction.

[Note 1]
In the above description, the case where the modified discrete cosine transform (MDCT) is used as the time frequency conversion is illustrated for the purpose of concrete description, but of course, this is an example, and any other time frequency conversion may be used. Is possible.

[Note 2]
For simplification of description, the number of microphones is limited to 2 and the method of correcting an acoustic signal formed from two channel signals has been described above. However, in the present invention, the number of microphones is not limited to 2. That is, the technique described in each of the above-described embodiments can be applied to a multi-channel signal including three or more channel signals based on output signals of three or more microphones. As in the case where the signal restoration processing and the signal reduction processing are performed for each channel signal in each embodiment, when the technique described in each embodiment is applied to a multi-channel signal, basically, for each channel signal. Then, signal restoration processing and signal reduction processing may be performed.

In addition, when the technique described in each embodiment is applied to a multi-channel signal including first, second,..., Q-th channel signals, wind noise determination may be performed as follows ( q is an integer of 3 or more).
For example, two channel signals are selected from the first to q-th channel signals, and the selected two channel signals are handled as the L signal and the R signal described above in the respective embodiments. Similarly, the degree of wind noise influence is determined through correlation value calculation.
Or, for example, for each combination of two channel signals of the first to q-th channel signals, a correlation value representing a cross-correlation between the two channel signals is obtained for each combination, and the correlation obtained for each combination is obtained. The degree of influence of wind noise is determined based on the maximum value, average value, minimum value, and the like.
Alternatively, for example, a correlation value indicating a cross-correlation between three or more channel signals among the first to q-th channel signals is obtained, and the influence level of wind noise is determined based on the correlation value.

  Further, the above-described first to fourth modified signal processes can be applied to the multichannel signal.

  When attention is paid to the first embodiment, the first to qth channel signals forming the multichannel signal are given to the BPF 30. When the first modified signal processing is applied to the multi-channel signal, the influence level of the wind noise is determined through the correlation value calculation. If it is determined that the influence degree is relatively large, the BPF 30 is passed. The first to q-th channel signals may be averaged, and the output signal of the signal reduction unit 13 may be formed by the averaged channel signals.

  When the second modified signal processing is applied to the multi-channel signal, the influence level of the wind noise is determined through the correlation value calculation. If it is determined that the influence degree is relatively large, the first passing through the BPF 30 is determined. The signal levels of the q-th channel signals are compared with each other. Then, among the first to q-th channel signals that have passed through the BPF 30, the channel signal having the lowest signal level is specified as the minimum acoustic signal and each other channel signal is specified as the non-minimum acoustic signal, and all non-minimum acoustic signals The signal may be replaced with the minimum acoustic signal, and the output signal of the signal reduction unit 13 may be formed with each channel signal after replacement.

  When the third modified signal processing is applied to the multi-channel signal, the middle band is subdivided into n, and the influence level of wind noise is determined for each band subdivided into n through correlation value calculation. Then, the degree of influence is determined for each of the n subdivided bands, and the first to q-th channel signals on the frequency axis (that is, the band whose influence degree is determined to be relatively large) MDCT coefficients) may be averaged, and the output signal of the signal reduction unit 53 may be formed with each channel signal after the averaging.

  Even when the fourth modified signal processing is applied to the multi-channel signal, the middle band is subdivided into n pieces, and the influence level of wind noise is determined for each band subdivided into n pieces through correlation value calculation. Then, the degree of influence is determined for each of the n subdivided bands, and the first to q-th channel signals on the frequency axis (that is, the band whose influence degree is determined to be relatively large) The magnitude relationship between the MDCT coefficients) is evaluated, and the channel signal with the lowest signal level is specified as the minimum acoustic signal, and each other channel signal is specified as the non-minimum acoustic signal. Thereafter, all the non-minimum acoustic signals may be replaced with the minimum acoustic signals, and the output signal of the signal reduction unit 53 may be formed with each channel signal after replacement.

[Note 3]
The wind noise reduction unit 6a, 6b, or 6c shown in FIG. 3, FIG. 5, or FIG. 9 can be realized by hardware, software, or a combination of hardware and software. When the wind noise reduction unit (6a, 6b, or 6c) is configured using software, a block diagram of a part realized by software represents a functional block diagram of the part.

  Further, all or part of the functions realized by the wind noise reduction unit (6a, 6b or 6c) is described as a program, and the function is executed by executing the program on a program execution device (for example, a computer). You may make it implement | achieve all or one part.

[Note 4]
For example, it can be considered as follows.
The wind noise reduction device according to the present invention includes signal generation means for generating a low-band acoustic signal different from the low-band acoustic signal included in the input acoustic signal by signal restoration processing, and the wind noise reduction device 6a. Alternatively, in 6c, a signal generation unit is formed by each part referred to by reference numerals 23 to 25, and in the wind noise reduction device 6b, the restoration signal generation unit 61 functions as a signal generation unit (FIGS. 3, 5, and 5). 9).
The function of the 1st correction | amendment means which produces | generates a correction | amendment acoustic signal of a low band is responsible for the signal restoration part 12 in the wind noise reduction apparatus 6a or 6c, and the signal restoration part 54 in the noise reduction apparatus 6b.
The function of the 2nd correction | amendment means which produces | generates the correction | amendment acoustic signal of a middle band is responsible for the signal reduction part 13 in the wind noise reduction apparatus 6a, and the signal reduction part 53 takes charge in the noise reduction apparatus 6b or 6c. It can be considered that the second correction means includes or does not include the wind noise determination unit 11 (FIG. 3) or the wind noise determination unit 52 (FIGS. 5 and 9).

1 is an external perspective view of an imaging apparatus according to an embodiment of the present invention. It is a schematic block diagram showing the electric constitution of the imaging device of FIG. FIG. 3 is an internal block diagram of the wind noise reduction unit of FIG. 2 according to the first embodiment of the present invention. It is a figure showing the unit area of signal processing concerning the 1st example of the present invention. FIG. 6 is an internal block diagram of the wind noise reduction unit of FIG. 2 according to a second embodiment of the present invention. It is a figure which concerns on 2nd Example of this invention and shows the relationship of each flame | frame which is a process unit of an encoding. It is a frequency spectrum figure for demonstrating the restoration method of the signal by the restoration signal generation part of FIG. It is a figure for demonstrating the restoration method of the signal by the restoration signal production | generation part of FIG. It is an internal block diagram of the wind noise reduction part of FIG. 2 based on 3rd Example of this invention. FIG. 10 is an internal block diagram of an AAC encoder that can be used in combination with the wind noise reduction unit of FIG. 9. It is an internal block diagram of the conventional wind noise reduction apparatus. FIG. 6 is a conceptual diagram for explaining a first modified signal reduction process according to the first embodiment of the present invention. It is a conceptual diagram for demonstrating 2nd deformation | transformation signal reduction processing in 1st Example of this invention.

Explanation of symbols

MIC1, MIC2 Microphone 1 Imaging device 4 Audio signal processing unit 5 Recording medium 6, 6a, 6b, 6c Wind noise reduction unit 11, 52 Wind noise determination unit 51 Correlation value calculation unit (wind noise determination unit)
12, 54 Signal restoration unit 13, 53 Signal reduction unit

Claims (12)

In a wind noise reduction device that generates a corrected acoustic signal in which wind noise is reduced from an input acoustic signal,
If the predetermined bandwidth is a higher frequency than the first band single且a first band a predetermined band including a bandwidth of wind noise and a second zone,
The wind noise reduction device is
Based on an acoustic signal in a band having a higher frequency than the first band, included in the input acoustic signal,
The acoustic signal generated by the signal generation means, the signal generation means including signal generation means for generating the first band acoustic signal in which the wind noise is reduced compared to the first band acoustic signal included in the input acoustic signal First correcting means for generating a first corrected acoustic signal based on
Second correcting means for generating a second corrected acoustic signal that is an acoustic signal of the second band in which the wind noise is reduced by reducing a signal level of the acoustic signal of the second band of the input acoustic signal;
A wind noise reduction apparatus comprising: a corrected sound signal output unit that outputs the corrected sound signal based on the first and second corrected sound signals. The first correcting unit generates the first corrected acoustic signal based on the acoustic signal in the first band included in the input acoustic signal and the acoustic signal generated by the signal generating unit. The wind noise reduction device according to claim 1. The input acoustic signal consists of a plurality of channel signals,
The wind noise reduction device further includes:
Wind noise determination means for determining an influence degree of the wind noise on the input acoustic signal based on a cross-correlation between different channel signals of a predetermined band component of each channel signal including the wind noise band,
3. The wind noise reduction device according to claim 1, wherein the first correction unit generates the first corrected acoustic signal in accordance with a determination result of the wind noise determination unit. The input acoustic signal consists of a plurality of channel signals,
The wind noise reduction device further includes:
Wind noise determination means for determining an influence degree of the wind noise on the input acoustic signal based on a cross-correlation between different channel signals of a predetermined band component of each channel signal including the wind noise band,
3. The wind noise reduction device according to claim 1, wherein the second correction unit generates the second corrected acoustic signal in accordance with a determination result of the wind noise determination unit. The input acoustic signal is given to the wind noise reduction device as a signal on the frequency axis, and the input acoustic signal is composed of a plurality of channel signals,
The second correction means includes
Dividing the second band of the input acoustic signal into a plurality of element bands, and generating the second corrected acoustic signal on the frequency axis by reducing the signal level of the acoustic signal of each element band;
For each of the plurality of element bands, a cross-correlation of the acoustic signals of the element bands between different channel signals is obtained, and a reduction level of the signal level is determined for each element band based on each cross-correlation. The wind noise reduction device according to claim 1 or 2, wherein The input acoustic signal is given to the wind noise reduction device as a signal on the time axis, and the input acoustic signal is composed of a plurality of channel signals,
The first corrected acoustic signal generated by the first correction unit is a signal on a time axis,
The wind noise reduction device further includes:
Extraction means for extracting a component of a predetermined band not including the first band and including the second band from the input acoustic signal;
Time frequency conversion means for converting the signal format of the synthesized signal of the first corrected acoustic signal and the extraction signal of the extraction means from the time axis to the frequency axis,
The second correction means generates the second corrected acoustic signal on the frequency axis by reducing the signal level of the second band acoustic signal in the synthesized signal on the frequency axis,
The corrected acoustic signal output means includes an acoustic signal including the second corrected acoustic signal on the frequency axis obtained from the second correcting means and the first corrected acoustic signal on the frequency axis obtained from the time-frequency conversion means. The wind noise reduction device according to claim 1 or 2, wherein the corrected acoustic signal on the frequency axis is output based on The second correction means includes
Dividing the second band of the synthesized signal on the frequency axis into a plurality of element bands, and generating the second corrected acoustic signal on the frequency axis by reducing the signal level of the acoustic signal of each element band;
For each of the plurality of element bands, a cross-correlation of the acoustic signals of the element bands between different channel signals is obtained, and a reduction level of the signal level is determined for each element band based on each cross-correlation. The wind noise reduction device according to claim 6. The input acoustic signal consists of a plurality of channel signals,
The second correction means includes
Paying attention to all or part of the second band,
By averaging the sound signals of the band of interest for the plurality of channel signals included in the input sound signal, the signal level of the sound signal of the band of interest in the channel that is relatively affected by the wind noise is obtained. Reduce
The wind noise reduction device according to claim 1 or 2, wherein the second corrected acoustic signal is generated from a signal obtained by the averaging. The input acoustic signal consists of a plurality of channel signals,
The second correction means includes
Paying attention to all or part of the second band,
Among the acoustic signals of the band of interest for the plurality of channel signals included in the input acoustic signal, the acoustic signal having the minimum signal level is specified as the minimum acoustic signal and the other acoustic signals are specified as non-minimum acoustic signals,
By replacing the non-minimum acoustic signal with the minimum acoustic signal, the signal level of the acoustic signal in the band of interest in a channel that is relatively affected by the wind noise is reduced,
The wind noise reduction device according to claim 1 or 2, wherein the second corrected acoustic signal is generated from a signal obtained by the replacement. The wind noise reduction device according to any one of claims 1 to 9,
A recording device comprising: a microphone for generating the input acoustic signal for the wind noise reduction device. The wind noise reduction device according to any one of claims 1 to 9,
A microphone for generating the input acoustic signal for the wind noise reduction device;
An imaging apparatus comprising: an imaging unit for acquiring an image. In a wind noise reduction method for generating a corrected acoustic signal in which wind noise is reduced from an input acoustic signal,
If the predetermined bandwidth is a higher frequency than the first band single且a first band a predetermined band including a bandwidth of wind noise and a second zone,
The wind noise reduction method is
Based on an acoustic signal in a band having a higher frequency than the first band, included in the input acoustic signal,
A signal generation step of generating the first band acoustic signal in which the wind noise is reduced as compared to the first band acoustic signal included in the input acoustic signal;
A first correction step of generating a first corrected acoustic signal based on the acoustic signal generated in the signal generation step;
A second correction step of generating a second corrected acoustic signal that is an acoustic signal of the second band in which the wind noise is reduced by reducing a signal level of the acoustic signal of the second band of the input acoustic signal; Prepared,
A method for reducing wind noise, comprising generating the corrected acoustic signal based on the first and second corrected acoustic signals.

Images