JP4413043B2 - Periodic noise suppression method, periodic noise suppression device, periodic noise suppression program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the quality of target sound by suppressing periodic noise mixed into the target noise. <P>SOLUTION: A periodic noise suppression method includes a band division processing for dividing a first signal mixing the target sound and a noise signal and a second signal, mainly comprising a noise signal into a plurality of frequency bands respectively; a level calculating processing for finding a level at each frequency component of the second signal; a fundamental wave component detection processing for detecting the lowest frequency band, having a peak value on the basis of a level value at each frequency band calculated by the level calculating processing and detecting the fundamental wave components of periodic noise; a sharpness detection processing for detecting whether sharpness of the fundamental wave component detected by the fundamental wave component detection processing is a prescribed value or larger; and a peak suppression processing for suppressing the fundamental wave component included in the first signal and the signal of the frequency band of the integer times, when it is detected by the sharpness detection processing that sharpness is the prescribed value or larger. <P>COPYRIGHT: (C)2005,JPO&amp;NCIPI

Description

  The present invention uses a microphone signal in which a noise signal is mainly observed with respect to a signal in which periodic noise is superimposed on a target signal, and observes the periodicity of noise to suppress the periodic component. The present invention relates to a suppression method, apparatus, and program.

Conventionally, a technique for observing the period of periodic noise and suppressing the periodic component has been proposed (Non-Patent Document 1). In the prior art, first, the lowest peak frequency (hereinafter, fundamental frequency) of periodic noise is detected, and a frequency band that is an integral multiple of the peak frequency is detected.
J. S. Lim and A. V. Oppenheim, “Enhancement and bandwidth compression of noisy speech,” Proc. IEEE, Vol. 67, No. 12, pp. 1586-1604, 1979.

However, since the detection method of the fundamental wave component has been the method of detecting the frequency band with the largest power, some peak value is detected even at the moment when the periodic noise is interrupted and the stationary noise is sounding. In spite of the fact that it is an erroneous peak (a peak that does not actually need to be detected), in order to detect and suppress it, the signal was over-suppressed and the signal was distorted, resulting in sound quality degradation ( Problem 1).
Furthermore, peak detection is only meaningful when the noise signal has a certain level (when the noise signal is small, if the peak is detected and suppressed, it is over-suppressed and causes sound quality degradation). In the conventional method, the peak value is always detected while ignoring the signal size, so that even a small signal is suppressed, resulting in excessive suppression, resulting in deterioration of sound quality (Problem 2).

  It is an object of the present invention to propose a periodic noise suppression method, apparatus, and program that can eliminate these problems and stably suppress periodic noise.

In the first embodiment of the present invention, the first signal in which the target signal and the noise signal are mixed and the second signal mainly including the noise signal are divided into a plurality of frequency bands, and the second signal Based on the level calculation process for obtaining the level for each frequency component and the level value for each frequency band calculated by the level calculation process, the lowest frequency band with a peak value is detected, and the fundamental wave component of periodic noise is detected. Fundamental wave component detection processing, and peak sharpness detection processing for detecting whether the sharpness of the fundamental wave component detected in the fundamental wave component detection processing is a predetermined value or more,
Proposed is a periodic noise suppression method including a peak suppression process for suppressing a fundamental frequency component included in the first signal and a signal in an integer multiple of the fundamental frequency component when a peak sharpness of a predetermined value or more is detected in the peak sharpness detection process. .

Furthermore, in the second embodiment of the present invention, in addition to the periodic noise suppression method described above, the power of the fundamental wave component is calculated. When the power of the fundamental wave component is equal to or greater than a predetermined value, the fundamental wave component and an integral multiple thereof are calculated. We propose a periodic noise suppression method that suppresses the frequency domain.
The third embodiment of the present invention proposes that the peak suppression process described above can arbitrarily set the suppression amount.

According to the periodic noise suppression method proposed in the first embodiment of the present invention, as a method for detecting the fundamental wave component, not only the peak size but also the peak sharpness is not less than a predetermined value. Since stationary noise is not detected as periodic noise, and stationary noise is erroneously detected as periodic noise and no suppression operation is performed, excessive suppression can be prevented.
Further, according to the second embodiment of the present invention, since the condition of the suppression operation is that the power of the fundamental wave component is a certain value or more, the suppression operation is not performed when the noise is small. Therefore, excessive suppression can be prevented and deterioration of sound quality can be avoided.

  Further, according to the third embodiment of the present invention, in the peak suppression processing, the suppression amount for suppressing the peak can be set from 0 to an arbitrary value and can be changed. As a result, when the sound quality is important, the suppression amount is reduced. When the S / N ratio is emphasized, the suppression amount is increased, for example, by changing it to zero, an appropriate target signal according to the purpose, for example, a voice signal can be obtained.

A first embodiment of the present invention will be described with reference to FIG. The first embodiment corresponds to the periodic noise removal method proposed in claim 1 of the present invention and the periodic noise removal device proposed in claim 4 that operates using this periodic noise suppression method.
The periodic noise suppression apparatus 10 shown in FIG. 1 receives a mixed sound of the target sound of the target sound source S and the noise generated by the noise source N, outputs a first signal X ₁ (k), and a noise source. N is mainly sound receiving the noise generated, a microphone 1B for outputting a second signal _X 2 (k), these first signals _X 1 (k) and a second signal _X 2 (k) is band-divided, respectively The band dividing unit 2, the level calculating unit 3 that calculates the power level of each frequency component of the second signal X ₂ (k) divided by the band dividing unit 2, and the power level calculated by the level calculating unit 3 A case is shown in which the fundamental component detecting means 4 for detecting the fundamental component having the peak at the lowest frequency, the peak sharpness detecting means 5 and the peak suppressing means 6 are configured.

The band dividing means 2 can be composed of, for example, a Fourier transform means, and mainly analyzes the frequency of a noise signal and divides it into a plurality of frequency bands. Each of the frequency-divided band components can be expressed as X ₂ (ω ₁ ) to X ₂ (ω _N ).
The level calculation means 3 calculates the power level of each band component X ₂ (ω ₁ ) to X ₂ (ω _N ) that has been divided into bands. The fundamental wave component detection means 4 detects the lowest frequency band having a peak among the power levels calculated by the level calculation means 3 as a fundamental wave component.
The peak sharpness detection means 5 calculates how sharp the peak is with respect to the fundamental frequency detected by the fundamental wave component detection means 4. An example of the sharpness calculation method is shown in FIGS. FIG. 2 shows an example of peak arrangement in the case of periodic noise. FIG. 3 shows an example of peak arrangement in the case of stationary noise. That is, in the case of periodic noise, the peak sharpness is high, and in the case of stationary noise, the sharpness is low. The difference between the peak value of the detected fundamental frequency and the average power of the entire search width is calculated as a method for calculating high and low sharpness. In the example of FIG. 2, the difference is d1, and in the case of FIG. 3, it is d2.

When the difference d1 or the difference d2 exceeds a certain value T, it is determined that the peak is sharp, and the peak frequency is transmitted to the peak suppressing means 6 as a peak (fundamental wave) to be detected.
In the case of the periodic noise shown in FIG. 2, since the peak is sharp, the difference d1 between the peak value and the average value exceeds a certain value. That is, d1> T is established. In the case of the stationary noise shown in FIG. 3, since the peak is dull, the difference d2 between the peak value and the average value does not exceed a certain value, and d2 <T is established. The sharpness of the peak value can be detected by such an operation principle. Only if it has a sharp peak (i.e., only if d1> T is satisfied), the peak frequency omega _P of the subsequent peak suppression unit 6 navel is transmitted as a fundamental frequency of the periodic noise.

In the peak suppression means 6, the peak frequency ω _P that changes from moment to moment is set to a frequency having a peak ω _P as a fundamental frequency, and a minimum value of the integral multiple of the frequency (ω _P , 2ω _P , 3ω _P ... Nω _P ). The comb filter is configured by a digital filter, and the comb signal filters the input signal X ₁ (ω ₁ )... X ₁ (ω _N ) of the microphone 1A in which the target signal and the noise signal are mixed. Thus, the periodic noise mixed in the input signal X ₁ (ω ₁ )... X ₁ (ω _N ) is suppressed. FIG. 4 shows an example of frequency characteristics of the comb filter. When the noise is, for example, an automobile engine sound, the fundamental wave of the periodic noise fluctuates with time, so the frequency having the minimum value of the comb filter needs to fluctuate with time. In such a case, a time-varying comb filter (a digital filter whose frequency characteristics can be arbitrarily changed according to the filter function equation) is used, and the minimum value is made to follow the frequency ω _P of the fundamental wave of periodic noise. What is necessary is just to change the frequency which has.

FIG. 5 shows a second embodiment of the present invention. This embodiment corresponds to an example of the periodic noise suppression device proposed in claim 5 that operates using the periodic noise suppression method proposed in claim 2 of the present invention.
The difference between claim 2 and claim 1 is that the peak power calculation for calculating the power of the peak frequency ω _{P in} order to perform the control for detecting the peak frequency ω _P that is the fundamental wave of the periodic noise more precisely. The means 7 is added. The peak power calculation means 7 detects whether or not the detected peak frequency ω _P has a power equal to or greater than a certain value Z. Only when the power exceeds the threshold value Z, the detected peak frequency ω _P The peak frequency to be suppressed is transmitted to the peak suppression means 6.

The effects of providing the peak power calculating means 7 are as follows. For example, even if the detected peak frequency is sharp, if it is very small noise, if the peak is detected and suppressed, the suppression becomes excessive, causing distortion in the target signal. Inconvenience occurs. By inserting the peak power calculation means 7 at this point, since the suppression works only when the noise is of a certain magnitude, the excessive effect does not occur and the target signal (voice) is less likely to be distorted. Is obtained.
Next explained is the third embodiment of the invention. The third embodiment corresponds to an example of the periodic noise suppression apparatus proposed in claim 6 that operates using the periodic noise suppression method proposed in claim 3 of the present invention.

The feature of the third embodiment resides in that a function for controlling the amount of suppression of the peak frequency ω _{P to be} suppressed (and a frequency corresponding to a multiple thereof) is added to the peak suppressing means 6. FIG. 6A shows the characteristics of the comb filter described in the first and second embodiments, and FIG. 6B shows the characteristics of the comb filter described in the third embodiment. In the first and second embodiments, the signal is suppressed to 0 at the peak frequency (and its integral multiple frequency) (hereinafter referred to as suppression gain value 0). However, in the third embodiment, the suppression gain value of the peak frequency to be suppressed (and its integer multiple frequency) is changed to, for example, 0.33 or 0.5 (hereinafter, the suppression gain value 0.33 or 0.5). ). With this control, the amount of suppression can be adjusted. When sound quality is important, the suppression gain value is set conservatively, such as 0.33 or 0.5, and when it is desired to prioritize the SN ratio improvement amount, the suppression gain value Can be set to zero.

ただしここで、ａｂｓは絶対値を、ｐｉは円周率を表す。
以上説明した実施形態１、２及び３では説明を簡素化するために目的音と雑音とが混ざった音を受音するマイクロホン１Ａと、主に雑音を受音するマイクロホン１Ｂを各１本づつとした場合を説明したが、マイクロホンを複数設ける場合もある。この場合には、例えば、各マイクロホンの受音信号のレベルを測定し、そのレベルが最も大きなマイクロホンへの受音信号が最もＳＮ比が良い信号であるとして、その信号を処理に用いる。騒音信号が受音されるマイクロホンが複数ある場合も同様に、各マイクロホンへの受音信号のレベルを測定し、そのレベルが最も大きなマイクロホンへの受音信号を処理に用いる。 The change of the suppression gain value is realized by changing the expression forming the comb filter. An example of a filter function when the suppression gain value is 0 is shown in Expression (1), and an example of a filter function when the suppression gain value is 0.33 is shown in Expression (2).

Here, abs represents an absolute value, and pi represents a circular ratio.
In the first, second, and third embodiments described above, in order to simplify the description, a microphone 1A that receives a sound in which a target sound and noise are mixed and a microphone 1B that mainly receives noise are each one. However, there may be a case where a plurality of microphones are provided. In this case, for example, the level of the sound reception signal of each microphone is measured, and the signal is used for processing assuming that the sound reception signal to the microphone with the highest level is the signal with the best SN ratio. Similarly, when there are a plurality of microphones that receive a noise signal, the level of the received sound signal to each microphone is measured, and the received sound signal to the microphone with the highest level is used for processing.

  The periodic noise suppression method according to the present invention can be realized by causing a computer to execute a periodic noise suppression program described in a computer-readable program language. The periodic noise suppression program is recorded on a computer-readable magnetic disk or a recording medium such as a CD-ROM, installed in the computer from these recording media, and a CPU (central processing unit) provided in the computer Decrypted by and executed. As another example of the method of installing on a computer, there is a case of installing through a communication line.

  The periodic noise suppression method and apparatus according to the present invention is incorporated in a car navigation system that is equipped with a voice recognition device and enables voice input, suppresses engine sound mixed in the voice input to the car navigation system, and performs voice recognition. It can be used to improve the rate.

The block diagram for demonstrating 1st Embodiment of this invention. The characteristic curve figure for demonstrating operation | movement of the peak sharpness detection means used for embodiment shown in FIG. The characteristic curve figure for demonstrating operation | movement of the peak sharpness detection means used for embodiment shown in FIG. 1 similarly to FIG. The characteristic curve figure for demonstrating the characteristic of the peak suppression means used for embodiment shown in FIG. The block diagram for demonstrating 2nd Embodiment of this invention. FIG. 6A shows a characteristic curve for comparing the third embodiment of the present invention with the first and second embodiments, FIG. 6A shows the frequency characteristic of the peak suppression means used in the first and second embodiments, and FIG. 6B. Indicates the frequency characteristics of the peak suppression means used in the third embodiment.

Explanation of symbols

S Target sound source 5 Peak sharpness detection means N Noise source 6 Peak suppression means 1A Microphone that receives target sound and noise 7 Peak power calculation means 1B Microphone that mainly receives noise 10 Periodic noise suppression device 2 Band division Means 3 Level calculation means 4 Fundamental wave component detection means

Claims (7)

Band division processing for dividing each of the first signal in which the target signal and the noise signal are mixed and the second signal mainly including the noise signal into a plurality of frequency bands;
Level calculation processing for obtaining a level for each frequency component of the second signal;
Based on the level value for each frequency band calculated in the level calculation process, a fundamental wave component detection process for detecting the lowest frequency band having a peak value as a fundamental wave component;
A peak sharpness detection process for calculating the peak value of the detected fundamental frequency and the difference between the average power of the entire search width as the sharpness, and detecting whether the sharpness is a predetermined value or more,
A peak suppression process that suppresses the fundamental wave component included in the first signal and a signal in an integer multiple of the fundamental wave component when a peak sharpness of a predetermined value or more is detected in the peak sharpness detection process;
The periodic noise suppression method characterized by including. The periodic noise suppression method according to claim 1,
Including a process for calculating a peak value of the fundamental frequency, and when the peak sharpness detection process detects a sharpness of a predetermined value or more, and the peak value of the fundamental frequency is a predetermined value or more, the fundamental wave A periodic noise suppression method characterized by suppressing a component and an integer multiple of the frequency domain. In any one of the periodic noise suppression methods of Claim 1 or 2,
A periodic noise suppression method characterized in that the peak suppression process can arbitrarily set a suppression amount. A microphone that outputs a first signal in which target sound and noise are mixed;
A microphone that outputs a second signal mainly consisting of noise;
Band dividing means for dividing each of the output signals of the plurality of microphones into a plurality of frequency bands;
Level calculating means for calculating the level of each frequency component after the second signal is band-divided by the band dividing means;
Based on the level value for each frequency band calculated by the level calculation means, a fundamental wave component detection means for detecting the lowest frequency band having a peak value as a fundamental wave component;
Peak sharpness detection means for calculating the difference between the peak value of the fundamental wave component detected by the fundamental wave component detection means and the average power of the entire search width as sharpness and detecting that the sharpness is equal to or greater than a predetermined value When,
A peak suppression unit that suppresses the fundamental wave component included in the first signal and a signal in an integer multiple of the fundamental wave component when the peak sharpness detection unit detects a sharpness of a predetermined value or more;
A periodic noise suppressing device comprising: The periodic noise suppression device according to claim 4,
Calculating a peak value of the fundamental wave component, calculated power is provided a peak power calculating means for determining whether more than a predetermined value, the peak power calculating means when the peak value of the fundamental wave component is greater than a predetermined value A periodic noise suppression device that operates the peak suppression means when judged. In any one of the periodic noise suppression apparatuses of Claim 4 or 5,
The periodic noise suppression apparatus according to claim 1, wherein the peak suppression means includes setting means capable of arbitrarily setting a suppression amount. A periodic noise suppression program written in a computer-readable program language and causing the computer to execute one of the periodic noise suppression methods according to claim 1.

Images