JPH0643898A - Voice coding device - Google Patents

Abstract

PURPOSE:To improve tone deterioration when an information transmission speed is small by providing a quantization bit number storage part, utilizing it in a next analysis frame, providing a soundless detection part, storing a quantization bit number in a soundless part in the storage part and utilizing it in a sound part. CONSTITUTION:An input voice signal is predicted by a digital filter in adaptive, and a difference signal being a difference between the predicted voice signal and the input voice signal is generated, and the difference signal is quantized. Then, at the time of quantizing the difference signal, a storage part 48 storing the quantization bit number remaining without being allocated is provided, and then, is utilized in the next analysis frame. Further, the soundless detection part 49 judging whether an input voice is the sound part or the soundless part is provided, and when the difference signal is quantized from the soundless part, the quantization bit number allocated in adaptive at each partial section is kept to an absolute minimum, and the remaining quantization bit number is stored in the storage part 48, and is utilized in the sound part.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coder which utilizes an analog line or a digital line, predictively codes a voice signal input through the line, and compresses an information amount.

[0002]

2. Description of the Related Art Conventionally, as a high-efficiency coding method for speech,
For example, there is an APC-AB system and the like, but with the development of various voice services in recent years, it has been applied to a voice mail device or the like that stores and transmits voice data as encoded information, and is in a practical stage. A case will be described below in which a conventional voice encoding device connected to an analog telephone line is performing high-efficiency voice encoding by the APC-AB system.

FIG. 5 is a block diagram of a speech coder in a conventional speech coder. 1 is a line interface unit that is connected to an analog line to control the line and separate input / output, 2 is an A / D converter that converts an analog signal into a digital signal, and 3 is an input signal in a plurality of frequency bands A band splitting filter for splitting, 4 is a linear prediction coefficient extracting unit that extracts linear prediction coefficients that form a short-term prediction filter, and 5 is a short-term prediction filter of the linear prediction coefficient extracting unit 4, and uses a short-term prediction filter to output a speech signal based on proximity correlation. , A pitch detector for detecting the pitch of the voice signal, and a pitch gain 7 for deriving a pitch gain indicating the magnitude of the correlation of the voice signal with respect to the pitch period detected by the pitch detector 6. A derivation unit, 8 constitutes a long-term prediction filter using the pitch gain derived by the pitch gain derivation unit 7, and predicts a speech signal based on the pitch correlation. A predictor to perform, 9 is an adaptive predictor including a predictor 5 based on proximity correlation and a predictor 8 based on pitch correlation, and 10 is an output signal of the band-splitting filter 3 and a predictive signal output from the adaptive predictor 9. A residual signal quantizer that quantizes a residual signal that is a difference, 11 an inverse filter unit that derives a pseudo residual signal, and 12 equally divides the pitch period detected by the pitch detection unit into a plurality of sections. The partial interval setting unit, 13 is an average amplitude calculating unit that calculates the average amplitude of the pseudo residual signal derived by the inverse filter unit 11 for each partial interval, and 14 is according to the average amplitude calculated by the average amplitude calculating unit 13. A quantization bit number adaptation unit that determines the quantization bit number of the residual signal for each sub-interval, 15 is an auxiliary information quantizer that quantizes auxiliary parameters such as pitch information, linear prediction coefficient, and average amplitude, 16 An encoding unit that encodes the residual signal quantized by the residual signal quantizer 10 and the auxiliary parameter quantized by the auxiliary information quantizer 15 and outputs the encoded data as coded data, and 17 is an encoder in each band. A multiplexing unit that multiplexes the code data output from the unit 16 and a voice buffer 18 that stores the digital signal output from the multiplexing unit 17, and the above constitutes the encoding unit.

Next, the voice decoding unit will be described with reference to FIG. Reference numeral 19 is a code sequence separation unit that separates the coded data from the audio buffer 18 into each band, and 20 is coded data that is separated into each band by the code sequence separation unit 19 and is further quantized residual signal and pitch information. , An information sequence separation unit for separating auxiliary parameters such as linear prediction coefficient and average amplitude, 21 is an auxiliary information decoder for restoring encoded auxiliary information to original auxiliary information, and 22 is an auxiliary information decoder 2
1. A decoding bit number adaptation unit for clarifying the quantization bit number for each partial section of the quantized residual signal based on the average amplitude information restored in 1 A residual signal decoder that restores a residual signal, and 24 is a predictive synthesis that predicts an original signal from an output signal of the residual signal decoder 23 by a prediction filter configured by pitch information and auxiliary information of a linear prediction coefficient. 25 is a band synthesizing filter for merging output signals from the prediction synthesizing unit 24 for each band, 2
Reference numeral 6 is a D / A converter for converting a digital signal into an analog signal, and 27 is a line interface unit which is connected to an analog line and controls the line and separates input and output. The above constitutes a decoding unit.

The operation of the speech coding apparatus configured as above will be described below. First, an analog signal such as voice input to the line interface unit 1 through a telephone line or the like is converted into a digital signal by the A / D converter 2. The audio signal digitized by the A / D converter 2 is, for example, a QMF (QuadratureMi).
A plurality of bands, for example, band A (0 to 1 kH) is generated by the band division filter 3 including a color filter.
z), band B (1 to 2 kHz) and band C (2 to 4 kHz)
z) divided into three frequency bands.

The band A will be described below. The linear prediction coefficient extraction unit 4 extracts a linear prediction coefficient from the original signal in the band A to form, for example, an even-order fourth-order linear prediction filter. Linear prediction is performed in the predictor 5 based on the proximity correlation using this linear prediction filter. The predictor 5 based on the close correlation performs prediction using sample values relatively close to each other, for example, data of four samples past from the present time. The pitch detector 6 detects the pitch of the original signal in the band A, and the pitch gain deriving unit 7 derives the pitch gain for the detected pitch period. The predictor 8 based on the pitch correlation performs prediction between pitch periods, that is, pitch prediction for predicting a current voice signal from data of one pitch period past, by using a prediction filter composed of derived pitch gains. The signals predicted by the predictor 5 based on the proximity correlation and the predictor 8 based on the pitch correlation are combined and output from the adaptive prediction unit 9. The difference between the prediction signal output from the adaptive prediction unit 9 and the original signal in the band A is calculated, and the difference is input to the residual signal quantizer 10 and quantized.

Here, the number of quantization bits of the residual signal quantizer 10 is adapted according to the average amplitude of the pseudo residual signal.
Therefore, it is necessary to calculate the average amplitude of the pseudo residual signal before operating the adaptive prediction unit 9. First, the inverse filter unit 1
In 1, the pseudo residual signal is obtained from the original signal of band A.
The inside of the inverse filter unit 11 is a predictor 5 based on proximity correlation.
And the predictor 8 based on the pitch correlation, which is the same as the adaptive predictor 9.

Next, this pseudo residual signal is sent to the partial interval setting unit 1
Entered in 2. The partial section is obtained by equally dividing the pitch cycle Tp into a plurality of sections, for example, four sections, as shown in FIG. 7, and is set so that each section periodically repeats. First
The th partial section is detected based on the amplitude periodicity of the pseudo residual signal, and the position Td of the partial section is designated by the time length from the beginning of the frame to the beginning of the first partial section. The average amplitude calculation unit 13 calculates the average amplitude of the pseudo residual signal for each partial section, and the quantization bit number adaptation unit 14
Then, an operation of assigning the number of quantized bits to each partial section is performed from the average amplitude. The number of quantization bits R _{ij in the} i band j section is given by the following equation.

R _ij = R _f (R) + F (log ₂ U _ij ) (1) i = 1, ..., 3 j = 1, ..., 4, where R _f is the average bit rate of the i band, It is proportional to the total number of bits R assigned to the residual signal quantization.
Here, R is a constant uniquely determined by the information transmission rate, and F is a function proportional to the logarithm of the average amplitude U _ij of the i band j section pseudo residual signal.

FIG. 8 shows an example of the number of quantized bits for each partial section of the band A thus determined. The residual signal quantized by the residual signal quantizer 10 in accordance with the number of quantization bits is sent to the coding unit 16 and coded, and at the same time fed back to the adaptive prediction unit 9.

On the other hand, the auxiliary parameters such as the pitch information, the linear prediction coefficient and the average amplitude of the pseudo residual signal are also quantized by the auxiliary information quantizer 15 and then encoded by the encoder 16. The number of quantization bits at this time is uniquely determined by the information transmission speed. The code data output from the coding unit 16 of each band is multiplexed by the multiplexing unit 17 and stored in the voice buffer 18 as code data.

The above-mentioned signal processing process is performed in the band B and the band C.
However, although the same is true, since the long-term prediction filter and the sub-interval setting in the predictor 8 based on the pitch correlation use those configured in the band A, the pitch detection unit 6, the pitch gain derivation unit 7, and the sub-interval are set. The operation of the setting unit 12 is not performed. The above operation is the description of the encoding unit.

Next, an example of restoring a voice signal from the voice buffer 18 will be shown. First, the coded data from the audio buffer 18 is separated into a coded data series for each band by the code series separation unit 19. Further, the coded data is separated into coded data of the quantized residual signal and coded data of auxiliary parameters such as pitch information, linear prediction coefficient, and average amplitude in the information sequence separation unit 20 of each band.

The signal processing of band A will be described below. The coded data of the auxiliary parameters are input to the auxiliary information decoder 21 and restored to the original auxiliary parameters such as the pitch information, the linear prediction coefficient and the average amplitude. afterwards,
In the decoding bit number adaptation unit 22, the decoding bit number R _ij of the quantized residual signal is calculated from the restored average amplitude U _ij based on the equation (1), and based on this, the quantization residual number is calculated. The difference signal is restored to the original residual signal in the residual signal decoder 23. The predictive synthesis unit 24 is composed of the decoded pitch information, auxiliary information such as a linear prediction coefficient,
As shown in FIG. 9, it comprises a predictor 28 based on proximity correlation and a predictor 29 based on pitch correlation, and a predictor 5 based on proximity correlation and a predictor 8 based on pitch correlation in the adaptive predictor 9 of the encoding unit, respectively. Is the same as.
The decoded residual signal is input to the predictive synthesizer 24, and the audio signal for each band is reproduced. The above signal processing process is exactly the same for band B and band C.

The band synthesizing filter 25 fuses the audio signals restored for each band, converts the digital signal into an analog signal by the D / A converter 26, and outputs it as a reproduced signal to the line interface unit 27. .

[0016]

However, in the conventional speech coding apparatus, when performing the speech coding processing as described above, the auxiliary information per analysis frame and the number of quantization bits assigned to the residual signal are It is uniquely determined in proportion to the transmission rate. Therefore, when the information transmission rate is low, the number of quantization bits assigned to the auxiliary information and the residual signal also decreases. When the number of quantization bits assigned to the quantization of the auxiliary information decreases, the prediction accuracy of the adaptive prediction unit decreases and the quantization error of the residual signal increases. In addition, since the number of quantization bits assigned to the residual signal quantizer is small, there is a problem that the quantization noise is further increased and the sound quality is significantly deteriorated.

An object of the present invention is to solve the above problems and to provide a speech coding apparatus capable of ensuring a relatively good sound quality even when the information transmission rate is low.

[0018]

In order to achieve the above-mentioned object, the present invention (a) In general, since the quantization bit number is assigned to each sub-interval in the quantization bit number adaptation unit, the total is It is difficult to finely adjust the number of bits to be equal to the total number of bits R assigned to the quantization of the residual signal, and therefore, it is set to be equal to or less than R. The number of bits is ignored. (A) Further, in the non-voice part, the amplitude of the input signal is almost 0. Therefore, as is clear from the equation (1), the number of quantization bits allocated may be minimum. In view of the above facts, a quantization bit number saving unit is provided in the configuration of the conventional example,
Regarding (a), the number of extra quantization bits is stored for each analysis frame. Regarding (a),
A silence detector is provided to determine whether the input to the speech encoder is a voiceless section or a voiced section.In the case of no voice, the number of quantization bits assigned to the residual signal quantizer is minimized and the remaining The bit number of is stored in the quantization bit number saving unit.

[0019]

According to the present invention, with the above configuration, the number of quantized bits stored in the quantized bit number saving unit in the voiced part having a large average amplitude is converted into the total number of bits R allocated to the quantization of the residual signal. By performing the calculation of the equation (1) after adding to, the number of quantization bits is assigned to the residual signal quantizer more than that of the speech coding process in the conventional speech coding apparatus, and the residual signal is Can be quantized precisely.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of a voice encoding unit of a voice encoding apparatus according to this embodiment. Reference numeral 30 is a line interface unit connected to an analog line for controlling a line and separating input / output, and 31 is a line interface unit. An A / D converter that converts an analog signal into a digital signal, 32 is a band division filter that divides an input signal into a plurality of frequency bands, and 33 is a linear prediction coefficient that extracts a linear prediction coefficient that forms a short-time prediction filter The extraction unit 34 is a predictor that uses a short-term prediction filter to perform linear prediction of a speech signal based on proximity correlation.
Reference numeral 5 is a pitch detection unit that detects the pitch of the voice signal, 36 is a pitch gain derivation unit that derives a pitch gain that indicates the degree of correlation of the voice signal with respect to the pitch period detected by the pitch detection unit 35, and 37 is a pitch gain derivation unit. A predictor that forms a long-term prediction filter using the pitch gain derived from the unit 36 and predicts a speech signal based on the pitch correlation, and 38 is a proximity correlation-based predictor 34 and a pitch correlation-based predictor 37. An adaptive prediction unit consisting of 39
Is a residual signal quantizer for quantizing a residual signal which is the difference between the output signal of the band division filter 32 and the prediction signal output from the adaptive prediction unit 38, and 40 is an inverse filter unit for deriving a pseudo residual signal. , 41 is a partial interval setting unit that equally divides the pitch cycle detected by the pitch detecting unit 35 into a plurality of intervals, and 42 is the average amplitude of the pseudo residual signal derived by the inverse filter unit 40 for each partial interval. An average amplitude calculation unit 43 for calculating, a quantization bit number adaptation unit 43 for determining the quantization bit number of the residual signal for each of the partial intervals according to the average amplitude calculated by the average amplitude calculation unit 42, and 44 a pitch Auxiliary information quantizer for quantizing auxiliary parameters such as information, linear prediction coefficient and average amplitude, 45 is a residual signal quantized by residual signal quantizer 39 and quantized by auxiliary information quantizer 44. An encoding unit that encodes the generated auxiliary parameter and outputs the encoded data as coded data, 46 is a multiplexing unit that multiplexes the coded data output from the encoding unit 45 in each band, and 47 is output from the multiplexing unit 46. This is an audio buffer for storing a digital signal according to the related art. The above is exactly the same as the conventional example. Reference numeral 48 is a quantization bit number saving unit for storing the excess quantization bit number for each analysis frame.

The operation of the speech coding apparatus configured as above will be described below. First, in the initial state, the storage Rs of the quantization bit number saving unit 48 is 0. In this state, the line interface unit 3 is connected to the telephone line etc.
An analog signal such as voice input to 0 is converted into a digital signal by the A / D converter 31. The signal digitized by the A / D converter 31 has a band A, for example.
(0 to 1 kHz), band B (1 to 2 kHz) and band C
It is divided into three frequency bands (2 to 4 kHz).

The band A will be described below. A linear prediction coefficient is extracted by the linear prediction coefficient extraction unit 33 for the signal in the band A to form a linear prediction filter, and the linear prediction is performed by the predictor 34 based on the proximity correlation using the linear prediction filter. Also, the pitch detector 3
5 detects the pitch of the original signal in band A, and the pitch gain for the detected pitch period is the pitch gain deriving unit 36.
It is derived by. The predictor 37 based on the pitch correlation performs a prediction between pitch periods, that is, a pitch for predicting a current voice signal from data of one pitch period past by a prediction filter composed of the derived pitch gain. The signals predicted by the predictor 34 based on the close correlation and the predictor 37 based on the pitch correlation are combined to form an adaptive predictor 3
It is output from 8. The difference between the prediction signal output from the adaptive prediction unit 38 and the original signal in band A is calculated, and the difference signal is input to the residual signal quantizer 39 and quantized. The above is the same operation as the conventional example.

Further, the number of quantization bits of the residual signal quantizer 39 is adapted according to the average amplitude of the pseudo residual signal,
This derivation method is also the same as the conventional example, and the inverse filter unit 4
0, the average amplitude of the pseudo residual signal for each partial section is calculated by the average amplitude calculating section 42 through the partial section setting section 41. After that, the quantization bit number adaptation unit 43 performs the operation (1) equation for assigning the quantization bit number to each partial section from the average amplitude.

Here, R = ΣΣR _ij (2) does not always hold, and in many cases it does not hold. That is, R ≦ ΣΣR _ij (3) Therefore, the quantized bit number saving unit 48 stores the quantized bit number Rs represented by the equation (4).

Rs = R−ΣΣR _ij (4) Thereafter, in the same way as the conventional example, the quantization bit number adaptation unit 4
The residual signal quantized by the residual signal quantizer 39 according to the number of quantization bits determined as described above in 3 is sent to the encoding unit 45 and encoded, and at the same time, the adaptive prediction unit 38. Will be returned to.

On the other hand, the auxiliary parameters such as the pitch information, the linear prediction coefficient and the average amplitude of the pseudo residual signal are quantized in the auxiliary information quantizer 44 in the same manner as in the conventional example.
After that, it is encoded by the encoding unit 45. The number of quantization bits at this time is uniquely determined by the information transmission speed. The code data output from the coding unit 45 of each band is multiplexed by the multiplexing unit 46 and stored in the voice buffer 47 as code data.

The above-described signal processing process is performed in the band B and the band C.
However, although the same is true, since the long-term prediction filter and the sub-interval setting in the predictor 37 based on the pitch correlation are those configured by the band A, the pitch detection unit 35,
The operations of the pitch gain derivation unit 36 and the partial section setting unit 41 are not performed. The above operation is the description of the encoding unit.

Next, when the analysis frame shifts to the next analysis frame, only the quantization bit number adaptation unit 43 is different from the above operation. That is, the number of quantization bits R _fj is given by the following equation.

R _ij = R _f (R + Rs) + F (log ₂ U _ij ) (5) Here, R _f is the average bit rate of the i band, and the total quantization assigned when quantizing the residual signal. It is proportional to the sum of the number of bits R and the number of bits Rs stored in the quantization bit number saving unit 48. F is a function proportional to the logarithm of the average amplitude U _fj of the pseudo residual signal in the i band j section and is similar to the conventional example. FIG. 3 shows an example of the number of quantization bits for each partial section of the band A thus determined. As described above, in the first section in which the average amplitude is large and the influence on the quantization error is the largest, the number of quantization bits is assigned more than in the conventional example, and precise quantization is possible, and the error is reduced. The above operation is the description of the encoding unit.

The decoding operation is exactly the same as the conventional example. Further, by providing the band A with the silence detecting section 49 for judging whether the voice signal input from the line interface section 30 is a voiced section or a non-voiced section in the above configuration,
A larger number of savings bits Rs will be obtained. Only parts different from the above embodiment will be described with reference to FIG.

First, in the initial state, the number of quantized bits Rs stored in the quantized bit number saving unit 48 is zero. In this state, the silence detector 49 operates the average amplitude calculator 4
The average amplitude calculated in 2 is compared with a certain level T _0, and it is determined whether or not it is a non-voice part based on the magnitude relation. If it is determined that the sound is not silent, the same operation as in the above embodiment is performed. When it is determined that there is no sound, the quantization bit number adaptation unit 43 allocates the minimum number of bits using only the first term of the equation (1). Then, the remaining quantized bit number Rs is calculated by the equation (4) and is stored in the quantized bit number saving unit 48. The number of quantization bits Rs in this case is much larger than that in the above-described embodiment. This operation will be performed as long as the voiceless section continues. After that, if the silence detecting section 49 determines that the section is a voiced section, the same operation as in the above embodiment is performed. FIG. 4 shows an example of the number of quantization bits for each partial section of the band A determined in this way. In this way, the number of quantization bits in each section is increased by one bit, so that the quantization error is significantly suppressed.

[0033]

As described above, the present invention reduces the sound quality deterioration due to the quantization noise (quantization error) of the residual signal by providing the quantization bit number saving unit and using it in the next analysis frame. be able to. Furthermore, a silence detection unit is provided, and in the voiceless unit, only the minimum number of quantized bits is allocated when the residual signal is quantized, and all the rest is stored in the quantized bit number saving unit, and this saved bit number is stored in the voiced unit. By utilizing it, the sound quality deterioration can be remarkably improved especially when the information transmission speed is low.

Further, with respect to an irregular input speech signal represented by a consonant, even when the prediction accuracy in the adaptive prediction unit is lowered, the signal-to-quantization noise ratio (S / N ratio) can be improved. It is possible to provide a practically effective speech encoding device.

[Brief description of drawings]

FIG. 1 is a block diagram showing a speech coder of a speech coder according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a speech encoding unit of the same speech encoding apparatus.

FIG. 3 is a diagram showing the number of quantized bits for each partial section of a region determined by the speech encoding apparatus.

FIG. 4 is a diagram showing the number of quantized bits for each partial section of a region determined by the speech encoding apparatus.

FIG. 5 is a block diagram showing a speech encoding unit of a conventional speech encoding device.

FIG. 6 is a block diagram showing a speech decoding unit of the speech encoding apparatus.

FIG. 7 is an explanatory diagram of a partial interval setting method using a voice coding configuration.

FIG. 8 is a diagram showing the number of quantized bits for each partial section of a region determined by the speech encoding apparatus.

FIG. 9 is a concrete block diagram of a prediction synthesis unit in the speech decoding unit of the speech encoding device.

[Explanation of symbols]

 30 line interface unit 31 A / D converter 32 band division filter 33 linear prediction coefficient extraction unit 34 predictor based on proximity correlation 35 pitch detection unit 36 pitch gain derivation unit 37 predictor based on pitch correlation 38 adaptive prediction unit 39 residual Signal quantizer 40 Inverse filter unit 41 Partial interval setting unit 42 Average amplitude calculation unit 43 Quantization bit number adaptation unit 44 Auxiliary information quantizer 45 Encoding unit 46 Multiplexing unit 47 Speech buffer 48 Quantization bit number saving unit 49 silence detector

Claims (2)

[Claims] 1. A prediction means for adaptively predicting an input voice signal by a digital filter, and a pseudo residual signal corresponding to a residual signal which is a difference between the voice signal predicted by the predicting means and the input voice signal. Generating means, means for setting a partial interval by equally dividing the pitch period of the input speech signal, means for detecting the average amplitude of the pseudo residual signal for each partial interval, and the residual signal being a quantum signal. Means for adaptively allocating the number of quantization bits corresponding to the average amplitude for each of the sub-intervals, and the number of quantization bits remaining unallocated during the quantization of the residual signal A speech coding apparatus which is provided with a saving section for storing and uses it in the next analysis frame. 2. A means for determining whether an input voice signal is a voiced portion or a voiceless portion is provided, and when the voiceless portion is used, when the residual signal is quantized, it is adaptively assigned to each partial section. 2. The speech coding apparatus according to claim 1, wherein the number of quantization bits is kept to a necessary minimum, and the remaining number of quantization bits is stored in a saving section and used in a voiced section.