JP3510643B2 - Pitch period processing method for audio signal - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding / decoding device for compressing a voice signal with high efficiency, and more particularly to a pitch period processing method for a voice signal.

[0002]

2. Description of the Related Art CE is one of the audio encoding methods for encoding an audio signal at a low bit rate of about 4 kbps or less.
An LP (code excited linear prediction) method is known. In this CELP method, a prediction residual signal obtained by predicting an input voice signal with a prediction filter is quantized at the level of a synthetic voice signal weighted with perceptual weight. The speech coding method based on this method has an advantage that noise included in the output synthesized speech signal can be reduced, and can generate a good synthesized speech signal as compared with other methods, and thus has been widely used in recent years. Is coming.

Details of the CELP system are described in, for example, Document (1) M.
R. Schroeder and BSAtal, "Code-Excited Linear Pred
iction (CELP): high quality speech at very low bit r
ates, "Proc.IEEE Int.Conf.Acoust., Speech, Signal Pro
ceeding, pp.937-940 (1985).

Normally, in the CELP system, the synthesized speech signal obtained on the decoding device side is finally output through an adaptive post filter. This adaptive post filter is for improving the subjective quality, and by appropriately setting its parameters, the spectral envelope and pitch repetition of the synthesized speech signal are emphasized to suppress the noise contained in the synthesized speech signal. This is a filter that performs noise shaping. The parameters used in the adaptive post filter are pitch gain and pitch period, and in the CELP method, they are updated in time units called subframes having a time length of about 5 to 8 msec. Therefore, the values of the pitch gain and the pitch period are constant within one subframe.

By the way, it is known that the pitch gain and pitch period of an actual audio signal change continuously and relatively smoothly. However, in the configuration of the conventional adaptive post filter, the pitch gain and the pitch period, which are the parameter values, are discontinuous between subframes,
The quality of the output audio signal subjected to the filtering process deteriorates.

[0006]

As described above, the conventional adaptive post filter cannot sufficiently express the smooth changes in the pitch gain and the pitch period of the actual speech signal, and the parameter value discontinuity between the subframes. Occurs, and the quality of the output audio signal subjected to the filtering process deteriorates.

The present invention solves such a conventional problem, and by using a parameter value having no discontinuity between subframes, it is possible to obtain a high-quality output audio signal with less noise. An object is to provide a pitch cycle processing method.

[0008]

In order to solve the above-mentioned problems, a pitch period processing method for an audio signal according to the present invention is
The magnitude of the distance between the pitch period input in the immediately preceding first time unit and the pitch period input in the current first time unit is calculated, and the magnitude of the distance is greater than a predetermined threshold value. When it is smaller, the pitch period from the immediately preceding first time unit to the current first time unit is smoothed to obtain a pitch period that changes every second time unit shorter than the first time unit. , The pitch period for each second time unit, and when the distance magnitude is equal to or greater than the predetermined threshold value,
The pitch cycle input in the current first time unit is set as the pitch cycle for each of the second time units. Further, in the method for processing the pitch period of the audio signal of the present invention, the absolute value of the difference between the pitch period input in the immediately preceding first time unit and the pitch period input in the present first time unit is calculated. It is characterized in that the magnitude of the distance is obtained by using the above. Also, in the pitch period processing method for an audio signal of the present invention, the first time unit is a subframe unit and the second time unit is a sample unit.

[0009]

As described above, according to the present invention, the pitch gain and pitch period obtained in the first time unit are smoothed,
A smoothly changing pitch gain and pitch period are obtained in second and third time units shorter than the first time unit,
By using these as the parameter values of the adaptive post filter, it is possible to obtain a high-quality speech signal with less noise and smooth power change.

In particular, although the pitch gain is always smoothed, the smoothing is performed only when it is determined that the pitch cycle has continuity. Therefore, there is a problem caused by smoothing the pitch cycle even to a portion which originally has no continuity. Naturalness is prevented, and it is possible to obtain an audio signal with excellent naturalness. Further, according to the present invention, the pitch period input in the immediately preceding first time unit and the current
The pitch from the immediately preceding first time unit to the current first time unit when the magnitude of the distance to the input pitch period in the first time unit is smaller than a predetermined threshold value. Smoothing the pitch cycle can be performed by smoothing the cycle and obtaining the pitch cycle that changes for each second time unit shorter than the first time unit.

[0011]

Embodiments of the present invention will be described below with reference to the drawings. 3 and 4 show a CE according to an embodiment of the present invention.
It is a block diagram which shows the structure of the audio | voice encoding apparatus and audio | voice decoding apparatus of LP system.

In FIG. 3, a frame buffer 101 is a memory for accumulating an audio signal input to the input terminal 100 for one frame, and each block in FIG. 3 uses the frame buffer 101 for each frame or subframe. The following processing is performed.

First, for one frame of audio signal,
The prediction parameter calculation circuit 102 calculates a short-time prediction parameter using a known method (usually, 8 to 12 prediction parameters are calculated). The calculation method is described, for example, in Reference (2) "Digital Speech Processing" by Sadateru Furui. The calculated prediction parameter is input to the prediction parameter coding circuit 103. The prediction parameter coding circuit 103 codes the prediction parameter based on a predetermined number of quantized bits, outputs the code to the multiplexer 115, and outputs the decoded value P to the prediction filter 104, the perceptual weighting filter 105, and the influence signal. Creation circuit 107, long-term vector quantization circuit 109
And to the short-term vector quantization circuit 111.

The prediction filter 104 calculates the short-term prediction residual signal r from the input speech signal from the frame buffer 101 and the decoded value of the prediction parameter from the encoding circuit 103, and outputs it to the perceptual weighting filter 105.

The perceptual weighting filter 105 is a filter constructed based on the decoded value P of the prediction parameter and subtracts the signal x obtained by modifying the spectrum of the short-term prediction residual signal r.
Output to. The perceptual weighting filter 105 is for utilizing the auditory masking effect similarly to the weighting filter in the conventional example, and since the details thereof are described in the above-mentioned document (2), description thereof will be omitted.

The influence signal generating circuit 107 includes an adding circuit 11
The past weighted combined signal x from 2 and the decoded value P of the prediction parameter are input, and the past influence signal f is output. Specifically, the zero-input response of the perceptual weighting filter in which the past weighted combined signal x is used as the internal state of the filter is calculated, and is output as the influence signal f in units of preset subframes. As a typical value in a subframe at the time of 8 kHz sampling, about 40 samples obtained by dividing one frame (160 samples) into four are used. Further, the influence signal generation circuit 107, in the first subframe, combines the composite signal x of the previous frame generated based on the density pattern K determined in the previous frame.
To produce an influence signal f.

The subtraction circuit 106 outputs a signal u obtained by subtracting the past influence signal f from the perceptually weighted input signal x in subframe units to the subtraction circuit 108 and the long-term vector quantization circuit 109.

The long-term vector quantization circuit 109 uses the subtraction times.
Difference signal u from path 106, drive signal holding circuit 1 described below
Past drive signal ex from 10 and encoding circuit 103
Input the prediction parameter P from
At the subtraction circuit 108 and the quantized output signal u of the difference signal u.
Vector gain β (pitch gain
And index T (equivalent to pitch period)
To the multiplexer 115 and drive the long-term drive signal q.
It outputs to each of the motion signal holding circuits 110. At this time q
Between u and u = Q * h (1) (However, h is the impulse response of the perceptual weighting filter 105.
Answer, * represents a convolution operation). Sabufu
Vector gain β in ram units^(m) And index T
^(m) An example of a detailed method of obtaining (m is a subframe number)
Is shown below.

That is, the preset index T
A drive signal candidate of the current sub-frame is created using the gain β and the past drive signal, and this is input to the perceptual weighting filter to create a quantized signal candidate of the difference signal u. Optimal index T ^(m) so that the error with the signal candidate is minimized And optimal β ^(m) To decide. At this time, T ^(m) And β ^(m) Let q be a long-term drive signal of the current sub-frame created by using, and let q be a quantized output signal u of the difference signal u.

A method similar to this is, for example, PETER KROON
Et al. "A class of Analysis-by-Synthesic Predicat" published in Vol.SAC-6, pp.353-363, IEEE 1988.
iveCoders for High Quality Speech Coding at Rates
Paper entitled "Between 4.8 and 16 kbits / s" (Reference (3)
A known method similar to the method of obtaining the coefficient of the pitch predictor in the closed loop in () can be used. On the other hand, the subtraction circuit 108 outputs the difference signal v obtained by subtracting the quantized output signal u from the difference signal u to the short-term vector quantization circuit 111 in subframe units.

The short-term vector quantization circuit 111 receives the difference signal v and the prediction parameter P as input, and adds the quantized output signal v of the difference signal v in subframe units to the addition circuit 112.
To output the short-term drive signal y to the drive signal holding circuit 110. Here, there is a relation v = y * h (2) between v and y.

Along with this, the short-term vector quantization circuit 111 outputs the gain G and the code vector index I which form the short-term drive signal to the multiplexer 115. FIG. 5 shows a specific configuration example of the short-term vector quantization circuit 111.

In FIG. 5, a synthetic vector generation circuit 30
1 is the prediction parameter P and the code vector C ⁽ⁱ⁾ in the preset codebook 302 A pulse train is created from (i is an index of the code vector), and the pulse train is combined by a perceptual weighting filter generated from the prediction parameter P to obtain a combined vector V ^(i). Is generated and output to the inner product calculation circuit 304 and the power calculation circuit 305.

The codebook 302 stores the amplitude information of the density pulse, and a code vector C ⁽ⁱ⁾ predetermined for the index i. Is composed of a memory circuit or a vector generating circuit which can be drawn. Inner product calculation circuit 304
Is the difference signal V from the subtraction circuit 108 in FIG. 3 and the combined vector V ⁽ⁱ⁾ Inner product value of A and ⁽ⁱ⁾ Is output to the index selection circuit 306. The power calculation circuit 305 calculates the composite vector V ⁽ⁱ⁾ Power B ⁽ⁱ⁾ Is output to the index selection circuit 306. Index selection circuit 306
Then, the inner product value A ⁽ⁱ⁾ And power B ⁽ⁱ⁾ By using the evaluation value {A ⁽ⁱ⁾ } ² / B ⁽ⁱ⁾ (3)

The index I that maximizes is selected from the index candidates i, and the corresponding inner product value A
^(I) And power B ^(I) To output to the gain encoding circuit 307. The index selection circuit 306 further outputs the information of the index I to the codebook 302 and the multiplexer 115 in FIG. Gain encoding circuit 307
Then, the inner product value A ^(I) output from the index selection circuit 306 And power B ^(I) Ratio to A ^(I) / B ^(I) (4) is encoded by a predetermined method, and its gain information G is used as the short-term drive signal generation circuit 308 and the multiplexer 11 of FIG.
Output to 5.

The above equations (3) and (4) are, for example, International Conference on Acoustic, Speech by IM Trancoso et al.
and Signal Processing Paper "EFFICIENT PROCEDUR
ESFOR FINDING THE OPTIMUM INNOVATION IN STOCHATIC
The one proposed by "CODERS" (reference (4)) may be used.

The short-term drive signal generation circuit 308 has a code vector C ^(I) corresponding to the gain information G and the index I. Input as C ^(I) Is used to create a pulse train by a method similar to that in the composite vector generation circuit 301, and the pulse amplitude is multiplied by a value corresponding to the gain information G to generate the short-term drive signal y. This short-term drive signal y is output to the perceptual weighting filter 309 and the drive signal holding circuit 110 in FIG.

The perceptual weighting filter 309 is a filter having the same characteristics as the perceptual weighting filter 105 of FIG.
The quantized output v of the difference signal v, which is created based on the prediction parameter P, is input to the adding circuit 112 of FIG.

Returning to FIG. 3, the drive signal holding circuit 1
10 receives the long-term drive signal q output from the long-term vector quantization circuit 109 and the short-term drive signal y output from the short-term vector quantization circuit 111, and inputs the drive signal e _x in the sub-frame unit. Output to. Specifically, for example, a value obtained by adding q and y in units of subframes for each sample may be used as the drive signal e _x . Drive signal e _x of the current subframe, for use in long-term vector quantizer 109 as the past drive signal in the next subframe, is held in the buffer memory in the drive signal holding circuit 110.

The adder circuit 112 quantizes the output u ^{(m) for} each subframe. And v ^(m) Then, the sum signal x of the past influence signal f created in the current subframe is obtained and output to the influence signal creating circuit 107.

The information of each of the parameters G, I, β, T and P obtained as described above is multiplexed by the multiplexer 115 and transmitted from the output terminal 116 as a transmission code. Next, the speech decoding apparatus of FIG. 4 for decoding the code transmitted from the speech encoding apparatus of FIG. 3 will be described.

In FIG. 4, the code transmitted from the speech coding apparatus is input to the input terminal 200. The demultiplexer 201 separates the input code into the codes of the gain G, the index I, the gain β, the index T, and the prediction parameter P. Decoding circuits 203, 20
5, 206 and 207 decode the codes of the gain G, the index I, the gain β and the index T, and output them to the drive signal generation circuit 209. The other decoding circuit 208 decodes the code of the prediction parameter P and outputs it to the synthesis filter 210. The drive signal generation circuit 209 generates a drive signal by using each decoded parameter as an input.

The drive signal generation circuit 209 is specifically constructed as shown in FIG. 6, for example. 6, a codebook 500 has the same function as the codebook 302 shown in FIG. 5 in the speech coding apparatus, and the code vector C ^(I) corresponding to the index I. Is output to the short-term drive signal generation circuit 501. Short-term drive signal generation circuit 50
1 has the same function as the short-term drive signal generation circuit 308 shown in FIG. 5 in the speech coding apparatus, which receives the gain G as an input and outputs the short-term drive signal y to the addition circuit 506. The adder circuit 506 outputs the sum signal of the short-term drive signal y and the long-term drive signal q generated by the long-term drive signal generation circuit 502, that is, the drive signal e _x to the drive signal buffer 503 and the synthesis filter 210 of FIG.

The drive signal buffer 503 is an adder circuit 50.
The driving signal output from 6 is held up to the past by a predetermined number of samples from the present, and when the index T is input, the driving signal is output from the driving signal of T samples past in order by the number of samples corresponding to the subframe length. Has become.
The long-term drive signal generation circuit 502 receives the signal output from the drive signal buffer 503 based on the index T, multiplies the input signal by a gain β, and generates a long-term drive signal that repeats at a cycle of T samples, and an addition circuit. 5
It outputs to 06 in a sub-frame unit.

Returning to FIG. 4, the synthesis filter 210
Is a filter having a frequency characteristic opposite to that of the prediction filter 104 shown in FIG. 3 in the encoding device, and outputs a synthesized speech signal with the drive signal and the prediction parameter as inputs.

The adaptive post filter 211 uses the prediction parameter, pitch gain β and pitch period T to shape the spectrum of the synthesized speech signal output from the synthesis filter 210 so as to subjectively reduce noise, and then the buffer 212. Output to. FIG. 1 shows a specific configuration example of the adaptive post filter.

The adaptive post filter 20 shown in FIG. 1 comprises a pitch emphasis filter 31 and a formant emphasis filter 32.
And a filter unit 30 having a gain adjusting unit 33, and a control unit having a gain smoothing unit 21, a period smoothing unit 22, and a continuity determining unit 23 for controlling the filter unit 30.

The pitch emphasizing filter 31 clarifies the fine structure of the spectrum of the synthesized speech signal by emphasizing the repetition of the pitch of the synthesized speech signal output from the synthesis filter 210 in the speech decoding apparatus 10 of FIG. It is for. The transfer function of the pitch enhancement filter 31 is, for example, 1 / (1-εβz ^-T ) (5)

Can be used. Here, ε is 0.
It is known that a value of about 2 to 0.5 is suitable.
Moreover, as a condition for the pitch enhancement filter 31 to operate stably, it is necessary to adjust the values of ε and β so that | εβ | <1. On the other hand, the formant emphasis filter 32 is a filter for emphasizing the shape of the spectrum envelope of the audio signal, and its transfer function is, for example, A (z / r ₁ ) / A (z / r ₂ ) (6). You can Here, 0 <r ₁ <r ₂ <1, and A (z) represents a prediction filter composed of prediction parameters decoded on the side of the speech decoding apparatus.

The gain smoothing unit 21 inputs the pitch gain β in subframe units, performs smoothing between subframes, obtains a pitch gain b (n) that changes smoothly in each sample unit, and determines the pitch gain b (n). Input to the emphasis filter 31. As an example of the pitch gain smoothing method, the following simple filter processing can be used. b (n) = (1-α) · b (n-1) + α · β ^(m) (7)

Here, 0 <α <1, and α is a parameter for adjusting the strength of smoothing. β ^(m) Is Figure 4
Represents a pitch gain given to the m-th sub-frame, which is input from decoding circuit 206 in speech decoding apparatus 10 of FIG.

For example, let n = φ represent the position of the sample point at the final stage of the (m-1) th subframe. At this time, the pitch gains b (1), b (2), ... In smoothed samples in the m-th subframe are the pitch gains b (φ) in the last sample of the m−1th subframe. It is possible to obtain sequentially by using Eq. Examples of pitch gains before gain smoothing and after gain smoothing are shown in FIGS. 7 and 8.

Continuity determining section 23 receives pitch period T in subframe units input from decoding circuit 207 in speech decoding apparatus 10 in FIG. 3, and pitch period T is continuous between subframes. To determine whether interpolation is possible. An example of the determination method in the continuity determination unit 23 will be described with reference to the flowchart shown in FIG.

First, the pitch period of the immediately preceding sub-frame is T
It is set to F (step S1), and the pitch period T of the current subframe is input (step S2). Next, the distance DT = distance (T, TF) between the pitch period T of the current subframe and the pitch period TF of the immediately preceding subframe is calculated (step S3). This distance DT can be obtained by, for example, f
= 1 / T to represent distance by frequency level,
There is a method of using the absolute value of the difference between T and TF.

Next, the distance DT obtained in step S3
Is compared with a predetermined threshold value DTmax (step S4),
When DT is smaller than DTmax, that is, from the pitch period TF of the immediately preceding subframe to the pitch period T of the current subframe
When the change amount to is small, it is determined that the pitch cycle has changed smoothly between the immediately preceding subframe and the current subframe, and the determination value TSMOOTH = 'ON' is set (step S5). On the contrary, when the distance DT is larger than the threshold value DTmax,
TSMOOT as a judgment value for not smoothing T
Set H = 'OFF' (step S6).

Returning to FIG. 1, the period smoothing unit 22 is the decoding circuit 2 in the speech decoding apparatus 10 of FIG.
Pitch period T input from 07 in subframe units
And input the judgment value TSMOOTH from the continuity judgment unit 23,
When TSMOOTH = 'ON', the pitch period T is smoothed between subframes to obtain a pitch period t (n) that changes smoothly for each sample unit, and this is input to the pitch enhancement filter 32. In addition, the cycle smoothing unit 22
When TSMOOTH = 'OFF', the pitch period T is not smoothed, and the pitch period T given for each subframe is input to the pitch enhancement filter 31 as it is. Examples before and after performing such a smoothing process of the pitch period T are shown in FIGS. 9 and 10.

The pitch emphasizing filter 31 uses the pitch period t (n) and the pitch gain b (n) given for each input sample unit, and the following expression 1 / (1- εb (n) z ^{-t (n)} (8) A pitch filter represented by the transfer function of (8) is constructed to more smoothly emphasize the repetition of the pitch of the synthetic speech signal s (n).

In this way, since the degree of discontinuity between parameter frames in the adaptive post filter is significantly reduced, it is possible to obtain a high-quality output speech signal with much less noise than the output terminal 213 of FIG. You can

[0049]

As described above, according to the present invention, the pitch gain and pitch period obtained in a time unit such as a subframe are smoothed in the speech decoding apparatus, and a shorter time unit, for example, a sample unit is used. By obtaining the smoothly changing pitch gain and pitch period and using them as the parameter values of the adaptive post filter, it is possible to obtain a high-quality voice signal with less noise and smooth power change.

Further, the pitch gain is always smoothed, but the pitch period is checked for continuity, and smoothing is performed only when it is determined that there is continuity. It is possible to prevent the occurrence of unnaturalness due to smoothing of the cycle, and it is possible to obtain a high-quality and highly natural audio signal.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an adaptive post filter according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a processing procedure of a continuity determination unit in FIG.

FIG. 3 is a block diagram showing the configuration of a speech coding apparatus according to an embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a speech decoding apparatus according to the embodiment.

5 is a block diagram showing a detailed configuration of a short-term vector quantization circuit in FIG.

6 is a block diagram showing a detailed configuration of a drive signal generation circuit in FIG.

FIG. 7 is a diagram showing an example of pitch gain before gain smoothing processing according to the present invention.

FIG. 8 is a diagram showing an example of pitch gain after gain smoothing processing according to the present invention.

FIG. 9 is a diagram showing an example of a pitch period before a pitch period smoothing process according to the present invention.

FIG. 10 is a diagram showing an example of a pitch period after a pitch period smoothing process according to the present invention.

FIG. 11 is a block diagram showing a configuration of a conventional post filter.

[Explanation of symbols]

20 ... Adaptive post filter 21 ... Pitch gain smoothing section 22 ... Pitch cycle smoothing section 23 ... Continuity determination section 30 ... Filter section 31 ... Pitch enhancement filter 32 ... Formant enhancement filter 33 ... Gain adjustment section

   ─────────────────────────────────────────────────── ─── Continued front page       (56) References JP-A-53-143103 (JP, A)                 JP-A-3-245195 (JP, A)                 JP-A-3-245199 (JP, A)                 JP-A-63-27896 (JP, A)                 JP 61-150000 (JP, A)

Claims (3)

(57) [Claims] 1. A size of a distance between a pitch cycle input in the immediately preceding first time unit and a pitch cycle input in the current first time unit is obtained, and the size of the distance is predetermined. If it is smaller than the threshold value of, the pitch period from the immediately preceding first time unit to the current first time unit is smoothed, and changes every second time unit shorter than the first time unit. A pitch period is obtained and used as the pitch period for each of the second time units. When the magnitude of the distance is equal to or larger than the predetermined threshold value, the current first A pitch cycle processing method for an audio signal, wherein a pitch cycle input in one time unit is set as a pitch cycle for each second time unit. 2. The magnitude of the distance is determined by using the absolute value of the difference between the pitch cycle input in the immediately preceding first time unit and the pitch cycle input in the current first time unit. The method according to claim 1, wherein the pitch period processing method is performed. 3. The method according to claim 1, wherein the first time unit is a subframe unit and the second time unit is a sample unit.