JP3107620B2 - Audio coding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for compressing and encoding an audio signal.

[0002]

2. Description of the Related Art A high-efficiency encoding method for audio signals at a compression rate of 8 kbits / sec or less is based on a code-excited linear predictive encoding method by Atal et al. Or disclosed in the following document 1. Method. In this method, a voice signal is expressed by parameters of a vocal tract and parameters of an excitation source. Note that Literature 1 is characterized in that the parameters of the excitation source are vector-quantized by using a statistical codebook and an adaptive codebook.

Reference 1: NSJayant & JHChen, “Speech
Coding with Time-Varying Bit Allocations to Excit
ation and LPC parameters ", Proc, ICASSP-89, (1989) Normally, vocal tract parameters are multistage quantized by using vector scalar or by vector scalar combination. It discloses a reduction in the amount of quantization computation due to quantization.

Literature 2: "Recognition of plosives by two-stage vector quantization" IEICE Technical Report, SP90-
6 (May 25, 1990).

[0005]

On the other hand, in the code-excited linear predictive coding system, the vocal tract parameters are sent only once for a frame of a specific time length due to bit rate compression, and the vocal tract parameters are not transmitted. When parameters are used for subframes obtained by further dividing a frame, interpolation is used.

When the bit rate is about 8 kbit / sec, a frame length of about 20 msec can be used. Here, considering the state of the vocal tract, the speech signal is considered to keep the same statistical properties for about 20 ms, so that for compression encoding of about 8 kbit / sec, simply interpolating is particularly necessary. No quality degradation occurs.

However, when the bit rate is further compressed to about 4 kbit / sec, the frame length is set to 30 m.
It has to be set from seconds to about 40 ms, that is, longer than the length of the steady section of the voice.

As a result, the temporal resolution of the vocal tract parameters becomes insufficient, and simply interpolating the vocal tract parameters makes it impossible to accurately represent vocal tract information. For example, consonants are lost. For example, there is a problem that the quality of synthesized speech is deteriorated.

Therefore, the present invention provides a method of speech synthesis in which the frame length remains long and the changes in vocal tract conditions can be represented accurately enough for further compression of the bit rate. It is intended to provide an encoding method.

[0010]

SUMMARY OF THE INVENTION The present invention relates to a code-excited linear predictive encoding method. In this method, a code specifying a vocal tract parameter and an excitation source parameter is a code output.

In the present invention, the vocal tract parameters are quantized by one-stage vector or a plurality of stages in vector units to determine a quantized vocal tract main parameter which is a vocal tract parameter in frame units.

Further, by interpolating the preceding frame and the current frame in units of subframes using the quantized vocal tract main parameters, a quantized vocal tract interpolation parameter is created in subframe units, and the vocal tract parameter of the input speech signal is interpolated. The difference is calculated, and a vocal tract parameter error is obtained for each subframe.

Then, the vocal tract parameter error is vector-quantized or scalar-quantized for each sub-frame.
Determine the quantized vocal tract correction parameters.

[0014]

A code specifying a quantized vocal tract main parameter in frame units and a code specifying a quantized vocal tract correction parameter in subframe units are output codes of vocal tract parameters. , The vocal tract parameter main code, the vocal tract parameter correction code, and the excitation source parameter code are multiplexed and transmitted to the communication line.

[0015]

As a whole, vocal tract parameters are quantized in a plurality of stages to determine a vocal tract parameter code. In this specification, the parameter related to the quantization at the preceding stage (vector quantization) is distinguished as the main parameter of the quantized vocal tract, and the parameter related to the quantization at the subsequent stage (vector or scalar quantization) is distinguished as the quantization vocal tract correction parameter. , The advantages of multi-stage quantization, such as a reduction in the amount of vector calculations, are maintained.

In the present invention, an interpolation process is performed on a subframe basis between the preceding stage and the subsequent stage. That is,
A vocal tract parameter error between a vocal tract parameter (quantized vocal tract main parameter) of the current frame and a vocal tract parameter (quantized vocal tract interpolation parameter) interpolated in subframe units is obtained, and the error is quantized. Then, the vocal tract parameters (quantized vocal tract correction parameters) and their codes are determined in subframe units.

As described above, since the interpolation processing is performed at an intermediate stage, the dynamic range (vector space of vocal tract parameters) in the subsequent quantization is reduced, and even if information in subframe units is added, the compression of the bit rate is reduced. Becomes

A quantized vocal tract main parameter code, which is information corresponding to a frame, and a quantized vocal tract correction parameter code, which is information corresponding to a subframe, are output as vocal tract parameter codes for each frame. .

[0019]

FIG. 1 is a block diagram showing an encoder to which the present invention is applied. 101 is a vocal tract analyzer, 102 is a vocal tract parameter quantizer, 103 is an interpolator, and 104 is a parameter error quantizer. , 105 is an adaptive codebook, 106 is a statistical codebook, 107 is a synthesis filter, 108 is an error calculator, 109 is a minimum error selector, 110 is a multiplexer, 111 is a subtractor, 112 is an adder, 113 is An adder 114 is a subtractor. Next, the operation will be described with reference to FIG.

The A / D-converted input speech signal sequence is input in a specific frame length unit, and is analyzed by the vocal tract analyzer 101 to obtain vocal tract parameters.

The vocal tract parameters of the input speech signal are quantized by a vocal tract parameter quantizer 102, and the code of the quantized vocal tract main parameter (vocal tract parameter main code) is once in each frame by a multiplexer. Sent to 110.

The main parameter of the quantized vocal tract is the interpolator 1
At 03, the frame is further interpolated in subframe units and used as vocal tract interpolation parameters.

The vocal tract parameters used in the current subframe include errors due to quantization and interpolation.

Therefore, the error is obtained from the difference from the vocal tract parameter containing no error obtained as a result of the vocal tract analysis in the current subframe, and the error is quantized by the vocal tract correction parameter quantizer 104, and the vocal tract is quantized. The vocal tract parameter is determined, and the code (vocal tract parameter correction code) is sent to the multiplexer 110 once in each subframe.

FIG. 2 is a diagram for explaining the operation of the quantizers 102 and 104 and the interpolator 103, and shows an example in which one frame is composed of four subframes.

In FIG. 2, circles at both ends indicate quantized vocal tract main parameters in frame units, X indicates quantized vocal tract interpolation parameters in subframe units, and Δ indicates output from the adder 112. Is what you do.

The difference between the X mark and the Δ mark indicates the vocal tract correction parameters created by the subtractor 111 and the parameter error quantization 104.

Referring to FIG. 1, an adder 113 adds an adaptive excitation vector from the adaptive codebook 105 and a statistical excitation vector from the statistical codebook 106 to form an excitation signal vector. The synthesized speech signal is synthesized by the synthesis filter 117 using the vocal tract parameters (corrected vocal tract parameters) output from 112, and the error calculator 108 obtains an error from the input speech signal.

The multiplexer 110 was obtained by the above apparatus.
The vocal tract parameter code, the vocal tract correction parameter code, the adaptive excitation code, and the statistical excitation code are multiplexed and transmitted as a total code to the communication line.

FIG. 3 shows a block diagram of a decoder corresponding to the encoding of FIG.

In FIG. 3, a demultiplexer 201 converts a total code received from a communication line into a vocal tract parameter main code, a vocal tract parameter correction code, an adaptive excitation code,
And a statistical excitation code, which are sent to each device of the decoder.

The vocal tract parameter main code is inversely quantized by the vocal tract parameter inverse quantizer 202 to become a vocal tract main parameter. Further, the main vocal tract parameters are interpolated by the interpolator 203 for each subframe.

Next, the vocal tract correction parameter inverse quantizer 20
In step 4, the vocal tract parameter correction code is inversely quantized to obtain a vocal tract correction parameter, and the adder 208 corrects the vocal tract main parameter.

The excitation signal vector is calculated from the optimum adaptive excitation vector corresponding to the adaptive excitation code from the adaptive codebook 205 and the optimum statistical excitation vector corresponding to the statistical excitation code from the statistical codebook 206. Then, the reproduced voice output is synthesized by the synthesis filter 207 using the corrected vocal tract parameters.

[0035]

According to the present invention, vocal tract parameters can be updated in subframe units instead of frame units, and as a result, the temporal change of the vocal tract properties of a speech signal at low bits can be achieved. Can be configured so as to be able to sufficiently follow the above.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a speech encoder to which the present invention is applied;

FIG. 2 is an explanatory diagram of an interpolation and correction operation in FIG. 1;

FIG. 3 is a block diagram of a decoder corresponding to FIG. 1;

[Explanation of symbols]

 Reference Signs List 101 vocal tract analyzer 102 vocal tract parameter quantizer 103 interpolator 104 parameter error quantizer 105 adaptive codebook 106 statistical codebook 107 synthesis filter 108 error calculator 109 minimum error selector 110 multiplexer 111 subtractor 112 addition Unit 113 Adder 114 Subtractor

Continuation of front page (72) Inventor Yoshihiro Ariyama 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) References JP-A-5-6199 (JP, A) JP-A-4 −171500 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 11/00-13/08 G10L 19/00-21/06

Claims (1)

(57) [Claims] 1. A vocal tract parameter is obtained by performing linear prediction analysis on an input speech signal on a frame basis, and an excitation source parameter is obtained by evaluating an error between the input speech signal and a synthesized speech signal. In a speech coding method for quantizing the excitation source parameters and determining codes specifying the parameters, and expressing the speech signal by the codes, the quantization is performed by vector-quantizing the vocal tract parameters in frame units. Determining a vocal tract main parameter, and creating a quantized vocal tract interpolation parameter in subframe units by interpolating a previous frame and a current frame in subframe units using the quantized vocal tract main parameter; Calculating the difference between the vocal tract parameters and the quantized vocal tract interpolation parameters in subframe units,
Calculating a vocal tract parameter error in subframe units, quantizing the vocal tract parameter error in subframe units, determining a quantized vocal tract correction parameter, and specifying the quantized vocal tract main parameter in frame units a vocal tract parameter correction code specifying the quantized vocal tract correction parameter in the vocal tract parameters main code and sub-frame and the codes of the vocal tract parameters, the vocal tract parameter correction and the vocal tract parameters main code
Communication by multiplexing the code and the excitation source parameter code
A speech coding method characterized by transmitting the signal to a line .