JPS634300A - Voice encoding method and apparatus - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a voice encoding method and an apparatus therefor, which can suppress audio signals by 1% to 4.8 Kb/s (kilobits/segment), have a low bit rate, and have few traits. The present invention relates to a speech encoding method and apparatus for high-quality encoding using high performance.

[Conventional technology]

Examples of methods for encoding audio signals with high quality at a low transmission pit rate of about 4.8 Kb/s include %Gan 59-272435 (Reference 1) and %Gan 6O-1789.
11 (Reference 2), etc., in a voiced section, the sound source signal of the frame is divided into one pitch section (representative section).
A method of expressing this pulse train as a pulse train and transmitting this pulse train is known. According to this method, it is possible to obtain good sound quality without impairing naturalness even with low Vibration and Trait of about 4.8 Kb/s.

FIG. 3 is an explanatory diagram of pulse processing showing an example of a conventional method for determining a pulse train of a representative section in a voiced section. In Figure 3, (a) is a 17-frame audio waveform, and (b) is a 17-frame audio waveform.
is an example in which the audio waveform of ta) is divided into subframes corresponding to the pitch period P' d, and one frame is divided into subframe sections using the pitch period P' d and the position of the th pulse gl. are doing. In addition, (C) is an example in which a pulse train is obtained for each subframe, and fd) is an example of the representative section found by searching and the pulse train in this representative section. Applicable. In this case, the representative section can be found by searching for a subframe section that can reproduce good audio over the entire frame or by selecting a section that includes pulses with large absolute values. As sound source information, the position of such a representative section, that is, the position of subframe (2) in the case of FIG. 3, and time information T of the start point of the subframe in the intra-frame trap are transmitted.

[Problem that the invention seeks to solve]

However, in this type of conventional method described above, a large amount of calculation is required to search for a pulse train or a representative section, and additionally, auxiliary information such as the position of the representative section and the starting point of a subframe within a frame is required. Therefore, if you try to implement hardware using a general-purpose signal processor etc. that is available on the desk market, the former reason would mean that it would be a national problem to implement it with a simple device configuration. 4. Disadvantages: 0. The purpose of the present invention is to eliminate the above-mentioned disadvantages, and to reduce the amount of calculations and simplify the configuration of the device. B. To provide a speech encoding method and apparatus capable of synthesizing high-quality speech even at low transmission and data rates of Kb/s.

[Means for solving problems]

The uninvented audio encoding method inputs a discrete audio signal on the transmitting side, extracts a spectral parameter representing a short-time spectrum envelope and a pitch buff meter representing a pitch from the audio signal, and encodes the data into a predetermined time interval. The sound source information representing the divided audio signal is displayed as a pulse train or a combination of noise and pulse train of one pitch interval of which the time interval is divided into a plurality of predetermined parts, and the information representing the sound source signal and the The pitch parameter and the front aa spectrum parameter are combined and output, and on the receiving side, based on the pitch parameter and the sound source information, if the sound source information is the pulse train, the driving sound source signal is restored based on this pulse train, and the sound source information is In the case of a combination of the front noise and the pulse train, the apparatus is configured to include means for restoring the drive sound source signal based on the sound source information and synthesizing the sound signal by using it in combination with the spectrum parameter.

The audio encoding device of the device of the present invention also includes a parameter calculation circuit that extracts and encodes a spectral parameter representing a short-time spectrum envelope and a pitch parameter representing a pitch from an input audio signal, and divides the input audio signal into predetermined time intervals. a drive signal calculation circuit that calculates a pulse train in the pitch section or a combination of noise and a pulse train based on the voice signal, the pitch parameter, and the spectrum parameter, and encodes the previous 6d sound source information.

A multiplexer circuit that combines and outputs the output code of the parameter calculation circuit and the output code of the drive signal calculation circuit.

Furthermore, the audio decoding device of the device of the present invention includes a video signal.

A code sequence in which a code representing a pitch parameter, a code representing a spectral parameter, and a code representing sound source information are combined is input, and the code representing the pitch parameter, the code representing the pitch parameter, and the code representing the sound source information are input. a demultiplexer circuit that separates and decodes;
Based on the decoded pitch parameter and the decoded sound source information, the pulse train in the pitch section is a sound (2@
A driving sound source restoration circuit that restores the driving sound source signal based on this pulse train when it is information, and generates a pulse train and noise based on this sound source information to restore the driving sound source when the combination of noise and pulse train is sound source information. and a synthesis filter circuit that synthesizes and outputs an audio signal based on the driving sound source signal and the decoded spectrum parameter.

〔Example〕

Next, a detailed explanation of the present invention will be given with reference to the drawings.
) Figure 1(b) is a diagram showing an embodiment of the speech encoding method according to the present invention and the transmitting side of the apparatus, and Figure 1(b) is a diagram showing an embodiment of the speech encoding method according to the present invention and the receiving side of the apparatus. It is a professional diagram showing one example.

The sending side shown in FIG. 1(a) has a buffer memory 110.
.

Pitch analysis circuit 130. a parameter calculation circuit 140,
Pitch encoding circuit 150. The parameter sign north ′kI
160. i/Pulse response parking circuit 170° Autocorrelation function calculation circuit 180 e' fi fitting is times'#! ! 200° Lunaplexer 260, etc., and 1-(b
) The receiving side shown in the figure includes a demultiplexer 290° decoding circuit 'Mt300. Noise memory 310. Pitch decoding circuit 32
0. and a parameter decoding circuit 330. Drive signal restoration circuit 3
40. It is configured to include an interpolation circuit 3359, a synthesis filter circuit 350, and the like.

The basic processing contents of the present invention performed by the transmitting side and the receiving side are as described above.In other words, the present invention utilizes the periodicity of the audio signal to adjust the predetermined pitch interval within the frame during the voiced interval. The pulse train in this pitch interval is used as a representative interval representing the frame to represent the sound source signal, and in the unvoiced interval, the sound source signal is represented by a combination of the pulse train and noise. In this case, in the voiced section, the position of the representative section is determined in advance and the pulse train is determined only for that section, thereby greatly reducing the number of computations that contribute to the υ pulse search and the search for the representative section. In addition, since auxiliary information such as information indicating the position of a representative section or information indicating the start point of a subframe within a frame becomes insecure, it is possible to increase the amount of information to be added to the pulse train, resulting in high quality with fewer calculations. It is possible to reproduce the sound of the computer, and with a compact device configuration, it is an actual machine of Doware.

As a method for determining the vibration and position of a pulse train, in addition to the method described in the above-mentioned document fl), (2), for example, ANALYSIS-by-Synthesis (ANALYSIS-by-Synthesis) is used.
-8YNTHE8I8:A-b-8) is known, and its details are described in 1A New Model Op LPC Excitement by B-5-ATAL et al.
Four Grodusi/Gnatural Saku/Dinog Speech A, Trok Bitley Two (A N
EW MOL）ELOFLPCEXCITATION
P (JB PruUDUC-ING NATUI
(,AL 80UNDINGSPEEC)IAT
A paper entitled L (, JW BIT RATES) (
PROC,i,C,A,8. S, P+, pp 614
~617.1982) (Reference 3) etc., and the method of finding the sound source by combining a pulse train and noise in a silent interval is described in detail in the method of (Reference 2). Now, in FIG. 1(a), the audio signal X(n) is input through the input terminal 100, and is stored in the buffer memory 110 by a preset number of pulls/pulls.

K /< The parameter calculation circuit 140 has a buffer memory 1
] An audio signal is input in units of a preset number of samples starting from 0, and parameters representing the spectral envelope of the audio signal are calculated. The autocorrelation method is a well-known method for calculating this parameter, which is the partial autocorrelation coefficient.
QUANTIZATION PROP
ERTIES OF TRANSMI8SIONP
ARAMETER8IN LINE PREDIC
- TIVE 8YSTEMS) (IEE
ETRAN8. A, S, S, P, pp 309-32
1゜1983) (Reference 4) Temple. Returning again to FIG. 1(a), the description of the embodiment will be continued.

K The i-th parameter outputted from the Huff meter calculation circuit 140 is outputted to the zK parameter encoding circuit 60 and quantized in advance.

The encoded signal is encoded based on the number of signals and outputted to the multi-grain t260 as a code JKi. Furthermore, the parameter encoding circuit 160 once decodes lKi and uses the parameter decoding Ki obtained from it to obtain a linear prediction coefficient value a'i, which is sent to the impulse response calculation circuit 170 and the weighting circuit 2.
The 0 pitch, chi analysis circuit 130 that outputs 0 to 0 calculates the pitch period P' d using the output of the buffer memo 9 circuit 110. 1 Real Time Ingrimenotice 7 Op Time Domain Harmony, Ri Scaling Op Speech (REAL
-TIME IMPLEMEN-TATION O
F TIME l)OMAIN HAI(, -M
(JNIC5CALING OF 5PEECH5
-IGNALS) (IEEE TRAN)
S, A, S, S, P6. pp, 258-272.198
3) (Reference 5) and the like can be used. In this case, it is determined whether the audio signal of the frame is voiced or unvoiced using an autocorrelation coefficient at a position far away from the pitch. If there is no voice, the pitch period is set to 0.

The pitch encoding circuit 150 performs multiplex encoding on the pitch period P'd using a predetermined number of quantization bits, and outputs it to the multi-blephr 260 as a code 1d. At this time, if there is no voice, a code corresponding to 0 is output. In addition, the impulse response calculation circuit 170 that outputs the decoded P'd to the drive signal calculation circuit 220 inputs the prediction coefficient Asahi a'i from the K parameter encoding circuit 160 and inputs the prediction coefficient Asahi a'i from the K parameter encoding circuit 160, and outputs the P'd obtained by decoding to the drive signal calculation circuit 220. The impulse response h representing the transfer function of
Calculate w(n). Here, hw(n) (7:)
Apart from that, the same method as the impulse response Pttn circuit 210 described in FIG. The response hw(n) is calculated by the autocorrelation function calculation circuit 180 and the mutual correlation function correction circuit 21.
Output to 0.

The autocorrelation function calculation circuit 180 inputs the impulse response hw(n) from the innopulse response calculation circuit 170, calculates the autocorrelation coefficient 几hh-, and calculates the rectangular signal fluctuation calculation time 1%220.
Output to. Here, to calculate Rhh-, if you keep it,
The same method as the autocorrelation function calculation circuit 180 described in document 6) can be used.0 The weighting circuit 200 calculates
(n), the prediction coefficient a'i from the parameter encoding circuit 160 is input, xtn) is multiplied by l, and the obtained xw(n) is output. For this calculation, for example, the weighting described in FIG.
The drive signal calculation circuit 220 first uses the pitch period P'd to determine whether the -C frame is voiced or unvoiced.

And as a sound source signal representing an audio signal, 7ffP.

In a frame, a pulse train in a predetermined pitch interval (representative interval) is calculated, and in a silent frame, a sound source signal is calculated by a combination of a pulse train and noise.

FIG. 2 is an explanatory diagram of pulse processing showing a representative section in a voiced frame in FIG. 1(a) and how to obtain a pulse train within the section. (a) shows the audio waveform of the frame. First, the frame is divided into subframes each having a pitch period P'd, as shown in divided subframes (b).

For this division, the worst nap frame mJ AT S of the previous frame (Nl-1) is stored, and subframe division can be performed by shifting this point by P'd. If the previous frame is silent, you can use Ike's excellent '73fi: method, such as dividing the frame into subframes from the leftmost point or dividing by finding the rising point of the signal. Colander. Next, as shown in pulse train (C) of representative section, a predetermined number of pulse trains are calculated for a predetermined pitch section. Here, the subframe section approximately in the center of the frame is used as the representative section (subframe section ■ in the figure), but other suitable methods may also be used. The calculation of the pulse train in the representative section is performed as follows in order to prevent pulses from being found adjacent to each other at subframe boundaries and to efficiently find the pulses. First, did you add both @KL samples in the subframe section (P'd+2L)? Calculate the cross-correlation function Rhx (e) for the nozzle (in the section of Nc in (b)). Here, L is the autocorrelation function calculation circuit 180
Indicates the maximum delay time of the self-phase@field number ahh (e) obtained by. For calculation of 几hx (to) and Rhh (e), the calculation method in the cross-correlation function calculation circuit 210 and autocorrelation function calculation circuit 180 in FIG. 1(a) of the above-mentioned (Reference 6) can be referred to. . A pulse train search is performed using Rhx(e) and 几hh-, and a predetermined number of pulse trains are found in the representative section. For pulse search, the first method of the above (Reference 2) is used.
The pulse search method described in the drive signal calculation circuit 220 in Figure (a) can be referred to. Depending on the size of the representative section obtained in this way and the way the representative section is divided, there is a risk that the pulse of the great cancellation may be lost, so the representative section is also good.

On the other hand, in the unvoiced 7 frames, the sound source signal is determined by a combination of a pulse train and noise. For this, the method described in the above-mentioned (Reference 2) can be referred to.

In addition to this method, it is also possible to divide the frame into several subframes (each of a certain section), obtain the pulse train for only one section, and represent the sound source signal with noise in the remaining subframe sections. The subframe section for obtaining the pulse train can be selected from a subframe section in which the original audio signal or the prediction residual signal predicted from the original audio signal has a large influence.

When a pulse train is input, the encoding circuit 230 encodes the amplitude and position of the pulse train, and outputs these codes to the multiplexer 260.

For example, the same 77-coding method as the encoding circuit 230 described in Document 6 can be used as the encoding method for the sequence.

When pulse train and noise information is input, the pulse train is encoded using the same method as described above, and the noise is encoded using a predetermined number of bits for image width and phase. and outputs the code to the multi-blefsa.

The multi-branch circuit 260 is a K-parameter encoding circuit]60 code Jki and a pitch encoding circuit 150 code l.
d and the code of the encoding circuit 230 are input, and these are combined and output from the transmission side output terminal 270.

This concludes the explanation of the transmission side of the speech encoding method according to the present invention.

Next, the pathology on the receiving side of the speech encoding method according to the present invention will be explained with reference to FIG. 1(b).

The demultiplexer 290 separates the code representing the K parameter, the code representing the pitch period, and the code representing the sound source information from among the codes inputted from the receiving side input terminal 280, and separates the code into a code representing the K parameter, a code representing the pitch period, and a code representing the sound source information. The decoding circuit 320° outputs to the decoding circuit 300.

The K-parameter decoding circuit 330 decodes the K-parameter and outputs the post-coded value 'i to the interpolation circuit 335. The pitch post-coding circuit 320 decodes the pitch period P'd and outputs the post-coded value 'i' to the interpolation circuit 335. It is output to the interpolation circuit 335.

The decoding circuit 300 decodes the sound source information and outputs it to the drive signal restoration circuit 340.

The drive signal restoration circuit 340 uses the pitch period decoded value P'd, and if it is a value other than O, determines that it is a voiced frame and a pulse train can be used as a sound source, and converts it to the drive signal calculation circuit on the transmission side. Divide the frame into subframes for each pitch period P'd using the same 75 method as in 220. Then, the received pulse information is divided into subframes for each subframe section represented by the position of a predetermined representative section within the frame. generate a pulse train using Next, the pulse train is interpolated using the pulse train of the first representative section and the pulse string of the adjacent frame to generate a sound source pulse train of one frame, restore the drive sound source signal, and output it to the synthesis filter circuit 350.

For this interpolation method, a method similar to the drive signal restoration circuit 340 described in the above (Reference 6) can be used.

- When a combination of a sword, a pulse train, and noise is used as a sound source, the same processing as the drive signal restoration circuit 340 described in (Reference 2) is performed to obtain a drive power signal and output it to the synthesis filter circuit 350.

The interpolation circuit 335 interpolates the decoded parameters for each pitch period, and outputs the interpolated parameters to the synthesis filter circuit 350.

The synthesis filter circuit 350 inputs the driving sound source signal and the interpolated parameters, and sends the signal to the synthesis filter circuit 25 on the transmission side.
The synthesized audio signal X (
n) is calculated and outputted from the receiving side output terminal 360 as well.

This concludes the explanation of the voice encoding method according to the present invention (!$111).

The embodiment described above is merely one embodiment of the present invention, and modifications thereof are also contemplated.

For example, the drive signal calculation circuit 220 uses a combination of a pulse train and noise as a sound source in order to better represent the voice of Haruna in the unvoiced section and to achieve a good transition between the unvoiced section and the voiced section. When representing a combination of a pulse train and noise, the number of pulses may be adaptively changed from 0 (that is, only noise) to several. In this case, it is necessary to transmit t* information representing the number of pulses (for example, about 2 bits per frame). In addition, as the n method for counting the pulse train, there are other methods other than the method described in this example. , you can use my unsophisticated method. For example, a method may be used in which the amplitude of a previously determined pulse is propagated each time one pulse is determined. For details of this method, see the paper entitled ``Study of source pulse accumulation method in multi-pulse-driven speech coding method'' by Ono et al. (Acoustical Society of Japan gI Volume 57, 1983) (Reference 7), etc. can do.

Another method for calculating the noise source is, for example, to generate a noise signal for each subframe, and to generate noise that minimizes the error power between the signal synthesized from the noise signal and the audio signal in the subframe section. The method of selection is known.

For more information on this method, please refer to Beenis Atal (B.
``1 Stroke Stuck Coding Op Speech Signals at Very Low Bit Rain'' (5TOCH13TICCODI) by S, ATAL) et al.
NG OF 5PEECH5IGNALSAT
VERY LOW BIT R, ATES) (PR, QCl, ICC84, pp, 1610
~1613.1984) (Reference 8). Another known method is to obtain the amplitude and phase of the noise source from the prediction residual signal obtained by predicting the speech signal. This method does not require synthesis of the speech signal, so can be reduced, but the quality will deteriorate.For details on this method, see Inuyama's 1Linear predictive analysis synthesis method using a driving sound source in which the residual is modeled with noise.
A paper entitled (Proceedings of the Acoustical Society of Japan, 1981)
165-166) (Reference 9).

In addition, when calculating the pulse train, the value of the K parameter was assumed to be constant within a frame (the characteristics of the filter synthesis filter do not change within a frame), but while the value of the K parameter was smoothly changed δ for each subframe, You may also ask for Specifically, the value of the K parameter is interpolated for each subframe using the parameter values of the previous and subsequent frames, this value is converted into a prediction coefficient, and the weighting is performed by the circuit 200.
The signal is output to the ear pulse response calculation circuit 170, and a cross-correlation function and an autocorrelation function are calculated using the ear pulse response of the representative section to obtain a pulse train. In this way, temporally smoother spectrum changes can be obtained, and higher quality speech can be synthesized.

In addition, the interpolation of the pulse train and the parameters may be performed in synchronization with the pitch period based on the typical pitch period, or may be interpolated in synchronization with the pitch period in a predetermined pitch period (for example,
Bi near the center of the frame.

Interpolation may be performed based on the

Further, the pitch period can also be subjected to the same processing as the parameters.

Methods of interpolating these parameters other than linear interpolation may also be considered. For example, for pulse trains and pitch periods, logarithmic interpolation etc. can be considered.Also, interpolation of parameters of a synthesis filter can be performed using, for example, a linear pre- 641 ), a method of interpolating a logarithmic cross section function, a format parameter, an autocorrelation function, etc. can also be used.

These specific methods are
.

1 speech analysis by K ATAL) et al.
Shi7se7su By Linear Pridi Kunyo 7 Og The Speech Waygu (SP-EECHANAL
YSIS AND 5YNTHE-5Is
BY LINEIL PREDIC'l0NO
F THE 5PEECHWAVE) and 111i1
．． paper (J, ACOUST, SOC, AM, p. 9)
．． 637-655.1971) (Reference 10).

Other methods can also be used to select the Saratra representative pitch section.

In this example, the parameters were analyzed and the sound source pulse train was calculated assuming that the frame length was constant, but it is also possible to make the frame length variable. In this case, the frame length can be shortened in the voice changing portion and lengthened in the constant portion, so that the transmission bit rate can be reduced.

Note that, as is well known in digital signal processing, the autocorrelation function can also be corrected from the power spectrum. Moreover, the cross-correlation function can also be calculated from the spatial vector.

Regarding these corresponding methods, please refer to the 1 digital signal processing “DIGITAL 8IG-NAL” by Nie Bui Otspenheim (A, ■, υPPENHEIM) et al.
A book titled “PROCESSING” (Reference 1])
It is explained in detail in Chapter 8 of the temple, so I will omit the explanation here.

All of the above can be easily implemented without detracting from the spirit of the present invention.

(Effects of the Invention) As described above, according to the present invention, a pulse train is searched for a predetermined pitch interval using the periodicity of the audio signal as the sound source signal in the voiced interval, Since the sound source signal of one frame is represented, the amount of calculation required for pulse search and representative section search can be significantly reduced.

Furthermore, since there is no need to worry about sending auxiliary information to indicate the position of a representative section or the starting point of a subframe within a frame, it is possible to allocate information to the pulse train accordingly. Therefore, very high-quality speech can be synthesized even at a low bit rate with a small number of calculations, and it has the advantage of being extremely easy to implement in hardware.

- In the silent section, the sound source signal is represented by a combination of pulse train and noise, so even the consonant waveforms and transient sound waveforms can be expressed extremely well. There is also the effect that the deterioration in sound quality is significantly reduced even if the voiced/unvoiced sound is not discriminated.

[Brief explanation of drawings]

FIG. 1(a) is a professional diagram showing the implementation of an embodiment of the speech encoding method and apparatus on the transmitting side according to the present invention;
) is a block diagram showing the configuration of an embodiment of the speech encoding method and apparatus on the receiving side according to the present invention, and FIG.
FIG. 3 is an explanatory diagram of pulse processing showing an example of a conventional method for determining a pulse train of a representative interval in a voiced interval. 110...Buffer memory, 130...
・Pi, Te analysis circuit, 140...K parameter calculation circuit, 150...Pitch encoding circuit, 160
...K /< parameter encoding circuit, 170...
...Impulse response calculation circuit, ]80...
Autocorrelation function calculation circuit, 220... Drive signal &
i-circuit, 225... Noise memory, 230...
... Code northbound r4.260 ... Multiplexer, 290 ... Demultiplexer, 300.
...Decoding circuit, 310...Noise mecha IJ
, 320...bi.

Claims (3)

[Claims] (1) On the transmitting side, a discrete audio signal is input, and a spectrum parameter representing a short-time spectrum envelope and a pitch parameter representing a pitch are extracted from the audio signal to represent the audio signal divided into predetermined time intervals. The sound source information for the sound source signal is expressed as a pulse train of one pitch interval of a plurality of predetermined divisions of the time interval or a combination of noise and a pulse train, and information representing the sound source signal, the pitch parameter, and the spectrum parameter are expressed. On the receiving side, based on the pitch parameter and the sound source information, if the sound source information is the pulse train, the driving sound source signal is restored based on this pulse train, and the sound source information is determined based on the combination of the noise and the pulse train. If the sound source information is used, the driving sound source signal is restored based on the sound source information, and the sound signal is synthesized by combining the sound source signal with the spectral parameter. (2) A parameter calculation circuit that extracts and encodes a spectral parameter representing a short-time spectral envelope and a pitch parameter representing a pitch from an input audio signal; a drive signal calculation circuit that encodes the sound source information by determining a pulse train in the pitch section or a combination of noise and pulse train based on the spectral parameter; an output code of the parameter calculation circuit; and an output of the drive signal calculation circuit. and a multiplexer circuit that combines and outputs a code. (3) Input a code sequence in which a code representing a pitch parameter, a code representing a spectrum parameter, and a code representing sound source information are combined, and separate and decode the code representing the pitch parameter and the code representing the sound source information. A demultiplexer circuit, based on the decoded pitch parameter and the decoded sound source information, restores the driving sound source signal based on the pulse train when the pulse train in the pitch section is sound source information, and combines the noise and the pulse train. In the case of sound source information, a driving sound source restoration circuit generates a pulse train and noise based on this sound source information to restore the driving sound source, and a sound signal is synthesized and output based on the driving sound source signal and the decoded spectral parameter. 1. A speech decoding device comprising a synthesis filter circuit.