JP2658438B2 - Audio coding method and apparatus - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Field of the Invention] The present invention relates to an audio signal having a low bit rate, particularly 4.8 kb / s.
The present invention relates to a speech coding system for performing high-quality coding with a relatively small amount of computation.

[Conventional technology]

As a method of encoding an audio signal at a low bit rate of about 4.8 kb / s, for example, an audio encoding method described in Japanese Patent Application No. 63-208201 (Document 1) is known. In this method, the transmitting side extracts a spectrum parameter representing a spectrum characteristic of a speech signal and a pitch parameter representing a pitch period from a speech signal of each frame, and uses the acoustic feature to divide the speech signal into a plurality of types (vowel,
Bursting, friction, etc.), and in the vowel section, the sound source signal of one frame is represented as follows by improved pitch interpolation. A multi-pulse is obtained for one pitch section (representative section) of a plurality of pitch sections obtained by dividing one frame for each pitch section. In other pitch sections of the same frame, an amplitude and phase correction coefficient for correcting the amplitude and phase of the multi-pulse in the representative section are obtained for each pitch section. Then, the amplitude and phase of the multi-pulse in the representative section, the amplitude in other pitch sections, the phase correction coefficient and spectrum, and the pitch parameter are transmitted. In addition, in a burst or transient section, a multipulse is obtained for the entire frame. In the frictional section, one type of noise signal is selected from a codebook including a predetermined type of noise signal so as to minimize the error power between the signal synthesized from the noise signal and the input audio signal. Calculate the optimal gain. Then, an index indicating the type of the noise signal and the gain are transmitted.

[Problems to be solved by the invention]

In the above-described conventional method, when the pitch period is correctly extracted without being affected by ambient noise or the like, it is possible to obtain a reproduced sound having good sound quality in a vowel section. However, if the pitch period is erroneously extracted by the speaker when the pitch period of the input voice signal is extremely long, when the pitch period is extremely short, or when ambient noise or the like is superimposed on the input voice signal. In the vowel section, the sound quality was considerably deteriorated. This deterioration is caused by the occurrence of phase distortion when an interpolation process is performed using an incorrect pitch period during pitch interpolation to restore a sound source signal. Therefore, it is important to accurately determine the pitch period in order to maintain good sound quality.

SUMMARY OF THE INVENTION It is an object of the present invention to solve the above-mentioned problems and to provide a speech encoding system with a good sound quality at about 4.8 kb / s with a relatively small amount of calculation.

[Means for solving the problem]

According to a first aspect of the present invention, a spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch are determined for each predetermined frame from an input discrete voice signal, and a sound source signal of the voice signal is efficiently used by using the pitch parameter. In a speech encoding method for representing and encoding, a plurality of types of different pitch periods are calculated, and for each of the pitch periods, the frame section is divided into small sections corresponding to the pitch period. In one of the sections, a multi-pulse composed of a plurality of pulses is obtained, and in another small section, a correction coefficient for correcting at least one of the amplitude and the phase of the multi-pulse is obtained to restore a sound source signal. An error power is obtained from a sound source signal and the audio signal, and a pitch cycle in which the error power is minimized is selected.

In the second invention, a spectrum parameter calculating means for obtaining and coding a spectrum parameter representing a spectrum envelope for each predetermined frame from an input discrete voice signal, and obtaining and coding a pitch parameter representing a pitch A pitch parameter calculation unit; a speech encoding device that efficiently represents and encodes the excitation signal of the speech signal using the pitch parameter; a plurality of types of pitch period calculation units that calculate different pitch periods; , The frame section is divided into small sections corresponding to the pitch period, and a multi-pulse composed of a plurality of pulses is obtained in one of the small sections. A correction coefficient for correcting at least one of the amplitude and the phase of the multi-pulse is obtained and encoded, and Sound source calculating means for restoring a signal, for each of the pitch periods, determining error power from the restored sound source signal and the audio signal, and selecting means for selecting a pitch cycle corresponding to a minimum value of the error power, The apparatus is characterized in that it has a spectrum parameter calculation means, an output means for combining and outputting codes output from the selection means and the sound source calculation means.

[Action]

The speech encoding system according to the present invention is particularly characterized by a pitch extraction method. N types of pitch periods (T _{1 to} T _N ) are calculated from audio signals in each frame section (for example, 20 ms) by using N different types of pitch extraction methods in parallel.

Here, as a method of extracting the pitch period, for example, a method based on a well-known autocorrelation function, a method using a modified covariance function, a method using an autocorrelation function of a prediction residual signal, a method using a cepstrum, and using an AMDF function There are methods. Each method is described in, for example, "Linear Prediction" by Markel, Gray et al. (Springer-Verlag, 1975)
And the "AC" by Rabiner et al.
omparative Performance Study of Several Pitch Dete
ction Algorithms "(IEEE Trans. ASSP, pp. 399-41
8,1976) (Reference 3) and Ramachandran “Pitch Predicti
on Filters in Speech Coding "(IEEE Trans.Acoust.Sp
eech and Signal Processing, pp. 467-478, 1989), which can be referred to, and the description thereof is omitted here.

For N types of pitch periods, the audio signal in the frame section is divided into subframe sections having a length equal to the pitch period, and a multipulse is once obtained in one subframe section. In other subframes, a gain and a phase correction coefficient for correcting the gain and phase of the multi-pulse are obtained, and the sound source signal of the frame is restored. The above processing is performed in parallel for each pitch cycle. Then, a pitch cycle that minimizes the error power with respect to the input voice is selected.

Next, the excitation signal is calculated and encoded by using the selected pitch period.
, The excitation signal calculation circuit, the encoding circuit, and the like.

〔Example〕

FIG. 1 is a block diagram showing a configuration of a speech encoding apparatus that implements the speech encoding method according to the first invention.

In the figure, on the transmission side, an audio signal is input from an input terminal 100, and an audio signal for one frame (for example, 20 ms) is stored in a buffer memory 110.

The LPS analysis circuit 130 performs a well-known LPC analysis from the audio signal of the frame as a parameter representing the spectral characteristic of the audio signal of the frame, and calculates a predetermined order P. For the specific calculation method, reference can be made to the K-parameter calculation circuit of the above-mentioned document 1. Note that the K parameter is the same as the PARCOR coefficient. Then the code l _k obtained by quantizing the number predetermined quantization bits of K parameters and outputs to the multiplexer 260, further the linear prediction coefficients and decodes it
a _i ′ (i = 1 to M) and output to the weighting circuit 200, the impulse response calculation circuit 170, and the synthesis filter 281. K
For the method of parameter encoding and the conversion from the K parameter to the linear prediction coefficient, reference can be made to the above-mentioned document 1.

In the pitch cycle calculation circuit 150, as shown in detail in FIG. 2, first, a plurality of types of different pitch cycle extraction circuits 151 ₁ to 151 ₁ .
151 _N are operated in parallel, and pitch periods T _{1 to} T _N of the audio signal for each frame are calculated. To calculate the pitch period,
For example, there are a method based on a well-known autocorrelation function, a method using a modified covariance function, a method using an autocorrelation function of a prediction residual signal, a method using a cepstrum, and a method using an AMDF function. For the respective methods, for example, the above-mentioned documents 2 to 4 can be referred to, and thus the description is omitted here.

In multi-pulse computing circuit 152 ₁ -152 _N, and divided into sub-frames per pitch cycle frame section using the pitch period T ₁ through T _N, with respect to one sub-frame (e.g., the head of the sub-frame of the frame) To obtain a predetermined number of multi-pulses. As a method of calculating the amplitude and phase of the multi-pulse, for example, Japanese Patent Application No. 57-231603 (Reference 5) can be referred to.

Then the correction coefficient calculation circuit 153 ₁ ~153 _N, using the multi-pulse, calculating the multi-pulse of the gain in the other subframes of the same frame, the phase correction coefficient for each sub-frame. Since a specific method of calculating the correction coefficient can be referred to the above-mentioned document 1, etc., the description is omitted here.

Next, the error power calculation circuits 154 _{1 to} 154 _N use the multi-pulse and the correction coefficient to calculate the excitation signal v
(N) is restored.

Here, c _j and _dj are gain and phase correction coefficients obtained in the subframe section j. g _i, m _i is the leading sub-frame i-th determined multi pulse amplitude, indicating the phase. T is the pitch period.

A signal is reproduced using the restored sound source signal, and the audibility weighting error power E _{w with} respect to the input audio signal is expressed by the following equation.

Here, Φ is a perceptual weighting input signal x _w represented by the following equation.
(N) and the impulse response of the auditory weighting synthesis filter
_hw (n).

R _hh (m) is an autocorrelation function of the impulse response h _w (n).

Here, when the gain correction coefficient is obtained for each sub-frame, the perceptual weighting error power E _wj for each sub-frame can be calculated as in the following equation.

Therefore, the perceptual weighting error power can be calculated by using the equations (2) and (3) without actually reproducing the signal. Since the first term of the equations (2) and (3) is a constant in the frame, the second term and the third term of the right side of the equations (2) and (3) are required to minimize the perceptual weighting error power. What is necessary is just to maximize the sum of the terms. Therefore, the pitch cycle selection circuit 155 selects and outputs the pitch cycle in which the sum of the second and third terms in the equations (2) and (3) is the largest from T _{1 to} T _N.

Returning to FIG. 1, encoder 160 outputs a code obtained by quantizing the selected pitch period with a predetermined number of bits to multiplexer 260, and decodes the code to a decoding pitch period T obtained by decoding the code. 'To the driving sound source restoring circuit 283, the sound source calculating circuit 220, and the correction coefficient calculating circuit 270.

The impulse response calculation circuit 170 calculates the linear prediction coefficient
Using a _i ′, the impulse response h _w (n) of the synthesis filter subjected to the perceptual weighting is calculated, and this is output to the auto-correlation function calculation circuit 180 and the cross-correlation function calculation circuit 210.

The autocorrelation function calculation circuit 180 calculates and outputs an autocorrelation function R _hh (n) of the impulse response up to a predetermined delay time. The operation of the impulse response calculation circuit 170 and the autocorrelation function calculation circuit 180 can be referred to the above-mentioned document 1.

The subtractor 190 subtracts the output of the synthesis filter 281 for one frame from the audio signal x (n) of the frame, and outputs the subtraction result to the weighting circuit 200.

The weighting circuit 200 passes the subtraction result through an auditory weighting filter whose impulse response is represented by _w (n), obtains a weighting signal _xw (n), and outputs the signal. For the weighting method, reference can be made to the aforementioned reference 1.

The cross-correlation function calculation circuit 210 inputs x _w (n) and h _w (n), calculates and outputs a cross-correlation function Φ _{xh up} to a predetermined delay time. This calculation method can be referred to the above-mentioned document 1.

Next, using the pitch period obtained by the encoder 160, the sound source calculation circuit 220 divides the frame into pitch sections having a length equal to the pitch period, based on the improved pitch interpolation in the vowel section of the input speech. The amplitude and the phase of a predetermined number of multi-pulses in one pitch section (representative section) are obtained. The specific method can be referred to the aforementioned document 1.

The pulse encoder 225 calculates the multi-pulse amplitude g of the representative section.
_i and phase _mi are coded with a predetermined number of bits and output to multiplexer 260, and are decoded and output to correction coefficient calculation circuit 270 and driving sound source restoration circuit 283.

The correction coefficient calculation circuit 270 calculates and outputs gain and phase correction coefficients in pitch sections other than the representative section.

The encoder 230 encodes the gain correction coefficient c _k and the phase correction coefficient d _k with a predetermined number of bits, and
Output to Furthermore, these are decoded and output to the drive signal restoration circuit 283.

The driving sound source restoring circuit 283 divides the frame into pitch intervals equal to the pitch period using the pitch period T ′, and generates a sound source signal d obtained by the multipulse in a representative interval.
(N) is generated, and in a pitch section other than the representative section, using the excitation signal of the representative section, the decoded gain correction coefficient, and the decoded phase correction coefficient, the excitation signal v (N) is restored.

v (n) = {c _k · d (n−T′−d _k ) + d (n)
(11) The synthesis filter 281 receives the reconstructed sound source signal v (n), inputs the linear prediction coefficient a _i ′, obtains a synthesized speech signal for one frame, and outputs an influence signal for the next frame. The calculation for one frame is output to the subtractor 190.
The calculation method of the influence signal can be referred to the specification of Japanese Patent Application No. 57-231605.

For a detailed calculation method in these circuits, reference can be made to the sound source signal calculation circuit, the gain and phase correction coefficient calculation circuit, the drive signal restoration circuit, and the like in Document 1 described above.

The multiplexer 260 includes a code indicating the amplitude and phase of the multi-pulse in the representative section, a code indicating the position in the frame of the representative section, a code of the pitch period, a code indicating the gain correction coefficient and the phase correction coefficient, and a code indicating the K parameter. Output in combination.

FIG. 3 is a block diagram showing a configuration of a speech encoding device embodying the second invention. 1 and 2 in FIG.
The components having the same reference numerals as those in the figures perform the same operations as those in FIGS. 1 and 2, and therefore description thereof will be omitted.

In the present invention, by the pitch period calculation circuits 151 _{1 to} 151 _N ,
N different pitch periods T _{1 to} T _N are extracted. Then, using these pitch periods, the sound source calculation circuits 220 _{1 to} 220 _N ,
The encoders 225 _{1 to} 225 _N , the correction coefficient calculation circuits 270 _{1 to} 270 _N , the encoders 230 _{1 to} 230 _N , and the error power calculation circuits 154 _{1 to} 154 _N are operated in parallel to perform N types of encoding.

In the selection circuit 300, when the error power is minimized, that is, when the equations (2) and (3) are used for the error power calculation, the encoding path that maximizes the second term is selected from N types of encoding paths. When using equation (10), an encoding path that maximizes the second and third terms is selected from N types of encoding paths.

The multiplexer 260 includes parameters of the encoding path selected by the selection circuit 300 (a code representing the position in the frame of the representative section, a code representing the amplitude and phase of the multipulse obtained in the representative section, a code of the pitch period, a gain, and a phase. A code representing the correction coefficient) and a code representing the K parameter are output in combination.

Each of the above-described embodiments is merely an example of the present invention,
Various modifications are also conceivable.

For example, in a pitch section other than the representative section, the gain correction coefficient _ck and the phase correction coefficient _dk are obtained and transmitted, but the decoded average pitch period T 'is interpolated for each pitch section using the adjacent pitch period. Accordingly, a configuration in which the phase correction coefficient is not transmitted can be adopted. The gain correction coefficient is not transmitted for each pitch section, but the value of the gain correction coefficient obtained for each pitch section is approximated by a least-square curve or a least-squares straight line, and the coefficient of the curve or the straight line is encoded. It is also possible to adopt a configuration in which transmission is performed. These methods can be used in any combination. With these configurations, the amount of information for transmitting correction information can be reduced.

As a phase correction coefficient, for example, “2.4 kbps Pitch Prediction Multi-pulse Speech C” by Ono and Ozawa et al.
As described in a paper entitled "oding" (Proc. ICASSPS4.9, 1988) (Reference 6), a linear phase term τ is obtained at the end of a frame, and this is distributed to each pitch section. Alternatively, the phase correction coefficient obtained for each pitch section may be approximated by a least-squares straight line or a least-squares curve, or the like. May be encoded and transmitted.

In the error power calculation circuit, the amount of calculation increases, but the error power between the input voice and the reproduced voice may be actually calculated. Specifically, by driving the synthesis filter seek reproduction signal (n) using the recovered source signal, it is also possible to calculate the error power E _w according to the following equation.

Also, as in Document 1, different sound source signals can be used depending on the characteristics of the audio signal of the frame. For example, the audio signal may be classified into vowel, nasal, friction, burst, and the like, and the configuration according to the present invention may be used in a vowel section.

The sound source signal is not limited to the improved pitch interpolation multi-pulse described in the present embodiment, but may be a pitch interpolation multi-pulse sound source, a pitch prediction multi-pulse sound source, or the like. For the specific method of obtaining the pitch-interpolated multi-pulse sound source, reference can be made to Japanese Patent Application No. 59-272435 (Reference 7). Also, how to find the pitch prediction multi-pulse sound source is
Japanese Patent Application No. 59-131925 (Reference 8) and Japanese Patent Application No. 63-14
Reference can be made to 7253 specification (Reference 9) and the like.

In addition, the K parameter was encoded as a spectrum parameter, and LPC analysis was used as an analysis method, but other well-known parameters, such as
LSP, LPC cepstrum, cepstrum, improved cepstrum, general cepstrum, mel cepstrum and the like can also be used. In addition, an optimal analysis method can be used for each parameter.

Further, in order to reduce the amount of calculation, the transmission side may omit the calculation of the influence signal. As a result, the driving sound source restoration circuit 283, the synthesis filter 281 and the subtractor
The 190 is unnecessary and the amount of calculation can be reduced, but the sound quality is reduced.

As is well known in the field of digital signal processing, the autocorrelation function is represented by a power spectrum on the frequency axis,
Since the cross-correlation function corresponds to the cross-power spectrum, it can be calculated from them. These calculations are described in “Digital Signal Pr
ocessing "(Prentice-Hall, 1975) (Reference 10).

〔The invention's effect〕

As described above, according to the present invention, a plurality of different types of pitch extraction methods are operated in parallel, a multipulse which is a sound source signal is once obtained, a final error power is calculated, and the best pitch period is obtained. Selected and coded,
Accurate pitch extraction is possible even for speakers that have ambient noise superimposed on the input speech or have difficulty extracting the pitch period.Encoding of good sound quality at a bit rate of about 4.8 kb / s There is a great effect that reproduced sound can be obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech coding apparatus for implementing a speech coding method according to the first invention, FIG. 2 is a diagram showing a configuration of a pitch period calculation circuit, and FIG. FIG. 2 is a block diagram illustrating a configuration of a speech encoding device that implements a speech encoding scheme. 110: buffer memory 130: LPC analysis circuit 150: pitch cycle calculation circuit 151 _{1 to} 151 _N ... pitch cycle extraction circuit 152 _{1 to} 152 _N ... multi-pulse calculation circuit 153 _{1 to} 153 _N ... correction coefficient calculation Circuits 154 _{1 to} 154 _N … Error power calculation circuit 155… Pitch period selection circuit 160, 225, 225 _{1 to} 225 _N , 170… Impulse response calculation circuit 180… Autocorrelation function calculation circuit 200… Weighting circuit 220, 220 _{1 to} 220 _N …… Sound source signal calculation circuit 230, 230 _{1 to} 230 _N … Encoder 260… Mux 270, 270 _{1 to} 270 _N …… Correction coefficient calculation circuit 281 …… Synthesis filter 283 …… Driving sound source restoration circuit 300 …… Selection circuit

Claims (2)

(57) [Claims] 1. A spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch are determined for each predetermined frame from an input discrete voice signal, and a sound source signal of the voice signal is efficiently used by using the pitch parameter. In the speech coding method of encoding and expressing in the following, a plurality of types of different pitch periods are calculated, and for each of the pitch periods, the frame section is divided into small sections corresponding to the pitch period. In one section, a multi-pulse composed of a plurality of pulses is obtained, and in another small section, a correction coefficient for correcting at least one of the amplitude and the phase of the multi-pulse is obtained to restore a sound source signal. A voice signal comprising: obtaining an error power from a signal and the voice signal; and selecting a pitch cycle in which the error power is minimized. Encoding method. 2. A spectrum parameter calculating means for obtaining and coding a spectrum parameter representing a spectrum envelope for each predetermined frame from an input discrete voice signal, and a pitch parameter for obtaining and coding a pitch parameter representing a pitch. Calculating means, in a speech coding apparatus for efficiently representing and coding the sound source signal of the speech signal using the pitch parameter, a plurality of types of pitch cycle calculating means for calculating different pitch cycles, and each of the pitch cycles Divides the frame section into small sections corresponding to the pitch period, obtains a multi-pulse composed of a plurality of pulses in one of the small sections, and obtains the multi-pulse in another small section. A correction coefficient for correcting at least one of the amplitude and the phase of the signal Sound source calculating means for restoring, for each of the pitch periods, obtaining error power from the restored sound source signal and the audio signal, selecting means for selecting a pitch cycle corresponding to the minimum value of the error power, the spectral parameter An audio coding apparatus, comprising: calculation means; said selection means; and output means for combining and outputting codes output from said sound source calculation means.