JPS61275900A - Voice information compression system - 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] [Table of Contents] Overview Industrial Field of Application Conventional Technology Problems to be Solved by the Invention Means for Solving the Problems Action Embodiments First Embodiment Second Embodiment Effect [Summary] In the APC-AB type voice storage response system, the encoder side determines whether there is a silent frame, generates and stores only silent frame information for the silent frame, and the decoder side detects silence from the input. When a frame is determined, a silent signal is generated and output for this frame, thereby reducing the amount of information.

Also, when the encoder side determines a silent frame, it generates only silent frame information, and increases and accumulates the amount of information of the voiced frame following this silent frame, and when the decoder side determines a silent frame from the input. A silent signal is generated and output for this frame, and the sound frame following this silent frame is decoded according to the increased amount of information, thereby improving the sound quality of the sound frame following the silent frame. Improve.

[Industrial application field]

The present invention relates to a method for compressing, storing, and reproducing audio information, and in particular uses an adaptive predictive coding method (APC-AB method) with adaptive bit allocation as a method for compressing and encoding audio information. This invention relates to an audio information compression method that aims to reduce the amount of information or improve the quality of reproduced audio by focusing on processing.

In a voice storage response system, it is necessary to compress and encode voice information and store it, and to reproduce the original voice signal from this compressed and encoded information. The compression encoding method for audio information used in this case is required to be able to reduce the amount of information N to be stored as much as possible, and to provide good quality of reproduced audio for the same amount of information.

[Conventional technology]

Conventionally, an adaptive differential PCM method (AD PCM method) is often used as an audio information compression method in such cases. The ADPCM system adaptively changes the quantization (encoding) allocated bits e according to the audio power, and when using this system, the audio is compressed into 32 Kbpa culm information l. It is known that when the information is stored and the original audio is reproduced from the information, a fairly good quality η audio can be obtained.

[Problem that the invention seeks to solve]

Audio normally contains a considerable proportion of silent parts, so when compressing and encoding audio, by processing silent parts separately from spoken parts, it is possible to improve the quality of reproduced audio. It is possible to reduce the amount of information to be stored while maintaining quality.

However, when encoding using the ADPCM method, new functions such as a correlator are required to calculate the audio signal power required to compress information during silent periods and to identify voiceless consonants. Therefore, there is a problem that the scale of the device increases.

[Means for solving problems]

As a compression coding method for audio, an adaptive predictive coding method (APC-AB method) with adaptive bit allocation is known.

The APC-AB method has more complex functions than the ADPCM method, and requires more processing time to process the captured audio signal by forming frames.
The advantage of this method is that it is possible to obtain audio with almost the same quality with an information cover of h (16 Kbpa).

When considering application to a voice storage response system, the long processing time is hardly a problem, but since it has a function to analyze and process the voice signals captured by forming frames, it is possible to process silent parts. This has the advantage that silent frames can be detected without requiring any special additional functionality. Furthermore, regarding the increase in circuit scale, when considering the entire system incorporating the silence detection function, no additional functions are required as described above, so there is no particular disadvantage compared to the ADPCM system.

The method of the present invention aims to reduce the noise or improve the quality of reproduced audio by performing processing during silent periods based on the APC-AB method.
This shows the 9 logical structure. In the first invention, 101 is an encoder that performs linear prediction on the audio input signal (determines and outputs prediction coefficients, and performs one prediction using the prediction coefficients to extract a residual signal. Then, the pitch period and pitch position 8 are determined by correlation calculation from the residual signal of
At the same time, it calculates and outputs the residual power and creates optimal output row/curved bit allocation information from this information, and predictively encodes the audio input signal according to this bit allocation information to generate a residual code. and output it.

102 is a decoder that creates optimal input decoding bit allocation information from the pitch period, pitch position, and residual power in the accumulated information, and calculates prediction coefficients and prediction coefficients in the accumulated information based on this bit allocation information. Predictive decoding is performed from the residual code.

Reference numeral 106 denotes an intra-frame power calculation circuit, which calculates the intra-frame power on the encoder side.

Reference numeral 104 denotes a frame determination circuit, which determines whether a voice frame or a silent frame exists on the decoder side.

Reference numeral 105 denotes a silent output generating means, which generates a silent signal on the decoder side.

In addition to the above, the second invention further includes the following means.

A silent frame number counter 106 counts the number of consecutive silent frames on the encoder side.

A silent frame number counter 107 counts the number of consecutive silent frames on the decoder side.

[Effect]

In the first invention, when the encoder side identifies a frame with low intra-frame power and low correlation of residual signals as a sound frame, only silent frame information is generated and stored, and sound frames are stored. do not. When the decoder side determines a silent frame, it generates a silent signal and outputs it during the frame period corresponding to the silent frame, thereby reproducing the audio signal and reducing the amount of information.

In the second invention, in response to the accumulation of silent frame information on the encoder side, a silent signal is generated on the decoder side, and is output during a frame period corresponding to the silent frame, and By increasing the amount of information and storing it on the encoder side for the voiced frame that follows the silent frame, and by decoding the voiced frame that follows the silent frame on the decoder side in accordance with the increased amount of information, Improves the sound quality of a silent frame followed by a sound frame.

〔Example〕

[First Embodiment] FIG. 2 shows a configuration on the encoder side in an embodiment of the system of the present invention. Input audio signal (e.g. 0~4KH
z) is converted into a digital signal by an analog-to-digital converter (A/D) 1, and is applied to a band division filter 2 to generate a low frequency signal (0 to 2 KHz).

It is divided into six bands: middle range (2 to 3 KHz) and high range (3 to 4 KHz). These three signals are each LP
C analysis circuit 5-1+ 3-2* Added to 5-5 and analyzed by linear prediction (LPC) analysis to extract prediction coefficients that are feature parameters of speech, and to generate predictions made up of the prediction coefficients. A residual signal is extracted from each band signal by the detector.

The residual power and correlation calculation circuit 4 includes each LPC analysis circuit 5
Correlation calculation is performed on the residual signals from -1 to 5-5,
Extract the pitch period and pitch position.

Based on the pitch period and pitch position, each band is divided into four subintervals. Then, the residual power for each subsection, the residual power for each band, and the total residual power are calculated and input to the adaptive bit allocation circuit 5. The adaptive bit allocation circuit 5 generates bit allocation information based on the pitch period, pitch position, and residual power information so that more bits are allocated to areas with more residual power.

The predictive encoders 6-1n 6-2+ 6-5 receive the low-frequency, middle-frequency, and high-frequency signals from the band division filter 2, and the LP
Using the prediction coefficients from the C analysis circuits 3-1+5-2v 5-5 and the bit allocation information from the adaptive bit allocation circuit 5, predictive coding is performed using the optimum bit e for each band. Generate compressed residual codes in each band. This signal, together with the prediction coefficients of each band and the pitch period, pitch position, and residual power information from the residual power and correlation calculation circuit 4, becomes the information to be stored in this method.

As described above, the basic encoding processing function of the normal APC-AB system is performed, but in the case of the present invention, the power within the frame is further calculated by the power within the frame calculation circuit 7. The residual power 8 and correlation calculation circuit 4 identifies a frame with low intra-frame power and low residual correlation as a silent frame, generates silent frame information, and outputs it as information to be stored.

As mentioned above, a normal kC audio signal includes a considerable proportion of silent sections. This silent section originally has no audio information, so if the same processing is applied to such a section, not only will the amount of information be redundant, but it may also amplify noise during playback. do not have. So the second
As shown in the figure, silent frame information is generated and stored using the intra-frame power calculation function and the residual correlation calculation function, and is a parameter that should be stored in the case of a normal voice frame. Prediction coefficients, pitch periods, pitch positions, and residual codes are not generated. Here, as the silent frame information, a code for distinguishing it from the encoded code in the frame when there is sound, and information about how many consecutive silent frames are to be generated. As a code for distinguishing between sound frames, there is a method of determining temperature using a several-bit code that cannot be used in the encoded code during sound, and a method of providing one bit as a control bit in advance for the sound frame. You could think so.

FIG. 6 explains the APC-AB encoded output according to the method of this embodiment, (al is the normal APC-AB encoded output omitted for comparison, and (b) is the method according to the present invention for the same information. This is the APC-AB encoded output of .

Figure 5 (t3 in al), +41, 16)
, (7) is a sound room, and (1) is shaded and omitted.
(21, f51 is a silent frame. Also, (b) [
In this figure, l is a silent frame (1).

Control information A corresponding to (2) and continuous silent frames [silence consisting of B) 1/-mu axis, +3'l, +4'l
are 44 f 7 frames t31, (sound frame corresponding to 41, ■ is silent frame (5) V, silent frame information consisting of corresponding control information A and continuous silent frame number B, (61, (7) are sound frames corresponding to sound frames (6) and (7), respectively.In this example, continuous silent frames [B is 010 in the case of ■, and 001 in the case of ■.

There are two possible formats for the silent frame control information, as described in al., but from the viewpoint of playback audio quality, information compression effect, difficulty of information playback, etc., one control bit is set for the sound frame as well. It is more advantageous to provide one. In other words, a bit for identifying whether there is a sound or no sound is set at the beginning of each frame, and when there is a sound frame V, information is added after that to reduce the number of bits allocated to the residual signal in the normal APC-AB system by one bit. For silent frames, information about a fixed number of bits indicating the number of consecutive silent frames is added after the identification bit to configure the frame.

Figure 4 shows an example of the structure of a sound frame and a silent frame.
(al is a sound frame, and after the 1-bit presence/silence identification bit (0 in this example) provided at the beginning, the APC
-By adding 255 bits of AB encoded output,
It is shown that one frame consists of 256 bits. In addition, (bl is a silent frame, and after the presence/silence identification bit (1 in this example) provided at the beginning, 3 bits of information indicating the number of consecutive silent frames are added, forming one frame consisting of 4 bits. It is shown that

Note that here, 6 bits are given as information on the number of consecutive silent frames, but if there are consecutive silent frames exceeding the maximum number of 7 that can be represented by this, the next silent frame 4 will be displayed again.
Assume that a new bit is generated.

FIG. 5 shows the configuration of the decoder 1 that decodes the information generated by the encoder shown in FIG. 2 and reproduces engineered speech.

In FIG. 5, the adaptive bit allocation circuit 11 generates bit allocation information from information on the pip period, pitch position t, and residual power in the same manner as in FIG. 2.

On the other hand, the residual codes of the low band, middle band, and still band are respectively given to predictive decoders 12-1 and 12-2. decrypt. The audio signals of each band are synthesized through a band synthesis filter 15 to generate reproduced audio.

As described above, the basic decoding process of the normal APC-AB method is performed, but in the method of the present invention, a frame determination circuit 14 is further provided, and a frame determination circuit 14 is provided at the beginning of the frame based on the accumulated information. When the presence/silence identification bits in the frame are read and it is determined to be a voice frame as in the case of (&) in Figure 4, the next 255 bits are regarded as information in the frame and the normal APC-AB method is used. Performs audio playback processing. Further, when it is determined that it is a silent frame as in the case of tb+ in FIG. 4, the following 3 bits are looked at and the number of consecutive silent frames is indulged, and a silent state is generated for a time corresponding to the number of frames. In the silent state, the reproduced sound is all (in order to eliminate the sense of incongruity in the case of silence, white noise from the white noise generator 15 is outputted via the band synthesis filter 13.

In the above-mentioned embodiments described with reference to FIGS. 2 to 5, the amount of information to be stored is reduced by processing during silent periods, but the amount of information to be stored remains unchanged and It is also possible to improve the sound quality.

That is, among the parameters to be encoded in the APC-AB method, the quality of reproduced audio is improved by increasing the code bit a of the prediction coefficient, residual power 3, and residual code. On the other hand, the pitch period and pitch position are information of integer values within a certain range (for example, 0 to 61), and increasing the encoded pitch la has no effect on improving sound quality.

Therefore, the quality of reproduced audio can be improved by increasing the amount of information for the voiced frame portion with respect to the margin of information amount created by compression Vζ of the silent frame.

[Second Embodiment] Figure 6 shows the configuration of the encoder side in another embodiment of the system of the present invention based on such a principle.
The same parts as in the figure are indicated by the same numbers, and 8 is a continuous silent frame number counter. The continuous silent frame number counter 8 counts the number of consecutive silent frames from the silent frame information of the residual power and correlation calculation circuit 4, controls the adaptive bit allocation circuit 5 based on the result, and controls the adaptive bit allocation circuit 5. Accordingly, the bit allocation information output to the predictive encoder 6-+t6-2+6-5 is changed. Predictive encoder”-1+62
When e6-5 receives changed bit allocation information, it performs IC encoding so that the amount of information becomes approximately twice that of the normal case, for example.

FIG. 7 shows an example of the encoded output in the embodiment shown in FIG. and the unequal case, (A
) is the normal APC-AB encoded output, and (Bl is the APC-AB encoded output of this embodiment).

In Figures 7(a) and (b), normal APC-AB
Silent frame in encoded output ((1 in case (a))
) to (3), (1) to f5) in case of (b), (9
)) is the sixth in the APC-AB encoded output of this embodiment.
Silent frame information (11 in the case of (a), I and II in the case of (b)) is encoded in the same manner as in the figure (bl) and is made up of control information A indicating a silent frame and the number B of consecutive silent frames.

On the other hand, in the case of a sound frame (fa) following a silent frame in the normal APC-AB encoded output, (4) to (6)
), (b) +61 ~ (s+, ([) )
In the APC-AB encoded output of this embodiment, it is encoded with approximately twice the amount of transmission, resulting in a voice frame (in the case of (a) (4)
In the case of ~(ψ, (b)) (ψ~(8).

(10')).

On the other hand, other sound frames ((a)
α and αη in the case of (b) are the same as those in the voiced frame ((～(11') in the case of (1)) without changing the amount of information.
, (1) in case (b).

In this way, when there are consecutive voice frames equal to the number of consecutive silent frames, the total amount of information does not change, but the amount of information in the voice frames increases, and there are fewer consecutive voice frames than the number of consecutive silent frames. When the amount of information in a voice frame increases, the amount of information in all frames simultaneously decreases. Therefore, as a whole, it is possible to improve reproduced audio with a slightly smaller amount of information than in the case of the normal APC-AB encoding method. A
In the PC-AB method, encoding is performed by adaptive bit allocation, so each parameter is
Encoding with double the code length can be performed relatively easily.

Table 1 shows a comparison of the breakdown of encoded pills in a normal voice frame and a voice frame having twice the amount of information according to this embodiment.

1 Table 8 shows an example of the configuration of a voice frame and a silent frame in the method of this embodiment, where (a) shows a normal voice frame, with a 1-bit voice/silence identification bit ( In this example, one frame consisting of 256 bits is formed by 0) and the following 255 APC-AB encoded bits. (bl indicates a voice frame with twice the amount of information, and is 501 by 1 bit of the presence/silence identification bit (0 in this example), followed by 500 bits of the APC-AB encoded output. One frame is formed of 11 bits. +61 indicates a silent frame, and 1 bit of the presence/silence identification bit (1 in this example) and the information of the number of consecutive silent frames @10 bits make up 1 frame of 11 bits. A frame is formed.

By adopting such a frame configuration, even when the number of consecutive silent frames is 1, which has the smallest information compression effect during silent periods, the 11 bits of a silent frame and the 501 bits of a sound frame, which has twice the amount of information, can be reduced. Since the sum is 512 bits, which is equal to two frames in the normal case, it can be seen that the quality of the sound frame can be improved by the amount of sympathy.

FIG. 9 shows the configuration of the decoder side in this embodiment system,
The same parts as in FIG. 5 are designated by the same numbers,
16 is a continuous silent frame number counter.

Further, FIG. 10 is a flowchart showing the procedure for audio reproduction in the decoder shown in FIG. 9.

During audio playback, the frame determination circuit 14 determines the first bit of the frame (step S1), and if it is 0 indicating a normal sound frame, it reads the following 255 bits (step 82) and determines the first bit of the frame (step S1). A sound frame having an amount of information is reproduced, and the first bit of the next frame is determined again.

When the frame first bit determination result in the frame determination circuit 14 is 1 indicating a silent frame, 10 bits indicating the number of consecutive silent frames are read (step S
5), set the continuous silent frame counter 16 (
Step S4): Generate silent frames a set number of times. Next, the first bit of the frame is determined (step S5), and if it is 1 indicating a silent frame, the number of consecutive silent frames is read again (step S6).
), if it is 0 indicating a voice frame, the counter 1
The number of consecutive silent frames is set again to 'C' (step 86), the 501 bits of the sound frame having twice the amount of information are read (step S7), and the sound frame is played back for one frame, and the counter is Subtract 16 by -1. Next, the first bit of the frame is determined (step sa), and if it is 1 indicating a silent frame, 10 bits indicating the number of consecutive frames are read (step S6).
, if the value is 0 indicating a sound frame, check whether the content of the counter 16 is 0 or not, and if it is not 0, read the next 501 bits again (step S7), and repeat the same procedure with the counter 16. Repeat until the content becomes 0. When the content of the counter 16 becomes 0, return to the beginning and repeat from the determination of the first bit of the frame (step S
1).

〔Effect of the invention〕

As explained above, according to the method of the present invention, APC-A
In the B-type voice storage response system, the amount of information can be reduced by performing processing during silent periods. Also, although the amount of information does not decrease much, it is possible to reproduce audio with better quality.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the basic configuration of the system of the present invention, FIG. 2 is a diagram showing the configuration of the encoder side in an embodiment of the present invention, and FIG. 3 is a diagram showing the encoded output in the embodiment of FIG. Figure 4 is a diagram showing a configuration example of a sound frame and a silent frame in the embodiment of Figure 2, Figure 5 is a diagram showing the configuration of the decoder side in an embodiment of the present invention, Figure gJ & 6 is a diagram showing the configuration of the encoder side in another embodiment of the present invention, FIG. 7 is a diagram explaining the encoded output in the embodiment of FIG. 6, and FIG. 8 is a diagram showing the configuration of the encoder side in the embodiment of FIG. FIG. 9 is a diagram showing a configuration of a decoder in another embodiment of the present invention. FIG. 10 is a diagram showing an example of the structure of a frame and a silent frame. FIG. It is a flowchart which shows the procedure. 1...Analog-digital converter (A/D) 2...
・Band division filter 5-+ *! l-2@3-5...LPC analysis circuit 4...Residual power and correlation calculation circuit 5...Adaptive bit allocation circuit<S-1v 6-2 y 6-3...Predictive encoder 7・
... Intra-frame power calculation circuit 8 ... Silent frame consecutive number counter 11 ... Adaptive bit allocation circuit 12-1, 12-2112-3 ... Predictive decoder 16
... Bandwidth synthesis filter 14 ... Frame judgment circuit 15 ... White noise generator

Claims (2)

[Claims] (1) Perform linear prediction on the audio input signal to obtain and output prediction coefficients, predict using the prediction coefficients, extract a residual signal, and calculate the pitch period and pitch position by correlation calculation from the residual signal. A code that calculates and outputs residual power, creates optimal output encoding bit allocation information from this information, and predictively encodes an audio input signal according to the bit allocation information to generate and output a residual code. vessel(
101) to accumulate each output of the encoder, create optimal input decoding bit allocation information from the pitch period, pitch position, and residual power in the accumulated information, and based on the bit allocation information. In a system for reproducing an audio signal, the system includes a decoder (102) that performs predictive decoding from the predictive coefficients and residual codes in the accumulated information, and the encoder (101) side calculates the power within the frame. An intra-frame power calculation circuit (103) is provided, and a frame in which the calculated intra-frame power is small and the correlation of the residual signal is small is identified as a silent frame, and only silent frame information is generated and stored, and the decoder (102) side includes a frame determination circuit (104) that determines whether there is a sound or no-sound frame, and a silent output generating means (105) that generates a silent signal.
What is claimed is: 1. An audio information compression method comprising: generating a silent state according to accumulated silent frame information and reproducing an audio signal. (2) Perform linear prediction on the audio input signal to obtain and output prediction coefficients, predict using the prediction coefficients, extract a residual signal, and calculate the pitch period and pitch position from the residual signal by performing a correlation calculation. A code that calculates and outputs residual power, creates optimal output encoding bit allocation information from this information, and predictively encodes an audio input signal according to the bit allocation information to generate and output a residual code. vessel(
101) to accumulate each output of the encoder, create optimal input decoding bit allocation information from the pitch period, pitch position, and residual power in the accumulated information, and based on the bit allocation information. In a system for reproducing an audio signal, the system includes a decoder (102) that performs predictive decoding from the predictive coefficients and residual codes in the accumulated information, and the encoder (101) side calculates the power within the frame. An intra-frame power calculation circuit (103) is provided, and a frame in which the calculated intra-frame power is small and the correlation of the residual signal is small is identified as a silent frame, and only silent frame information is generated and stored. A silent frame number counter (106) is provided on the side of the silent frame (101) to count the number of consecutive silent frames, and the output encoding bit allocation information is changed according to the counted value to determine whether the silent frame is followed by a silent frame. A frame determination circuit (104) that increases the amount of stored information and determines whether there is a voice or no voice frame on the decoder (102) side, and a silent output generating means (105) that generates a silent signal.
and a silent frame number counter (107) that counts the number of consecutive silent frames, and generates a silent state according to the accumulated silent frame information, and also generates a silent state according to the counted silent frame number. An audio information compression method characterized in that a voiced frame following a frame is decoded in accordance with an increased amount of information to reproduce an audio signal.