JP3537008B2 - Speech coding communication system and its transmission / reception device. - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coded communication system, and more particularly, to a technique for converting a speech signal into a digital signal format and transmitting the signal, thereby reducing the amount of information to reduce the transmission speed and narrow the bandwidth. Speech Encoding Communication System for Analysis and Synthesis System for Efficient Use of Transmission Bandwidth and Its Transmission
It relates to a receiving device .

[0002]

2. Description of the Related Art In low-speed speech coding for effective use of radio wave resources, a speech coding scheme of about several kbps is used.
Hybrid encoding typified by CELP (code excited linear prediction) encoding is the mainstream, and good reproduction audio quality is realized. FIG. 1 is a conceptual diagram of a main part of the CELP coding method. The CELP coding method is called code-driven LPC coding, and is a method of driving a speech synthesis filter using noise in a codebook as a sound source.
An optimum noise drive signal that minimizes an error signal between the input voice and the synthesized voice is extracted, and the code index and the gain information G are multiplexed as parameters and transmitted.

On the other hand, in the region of about 2400 bps, LP
C (linear predictive coding)
The analysis-synthesis type coding represented by the system is mainstream. FIG.
Is a block diagram of the LPC. The LPC system uses a single pulse having a noise or a pitch cycle as a driving sound source by switching based on unvoiced / voiced discrimination.
The C coefficient, pitch period and power information are encoded and transmitted.

[0004]

However, in the above-mentioned conventional configuration, as can be seen from the LPC block diagram, in analysis-synthesis coding, since the amount of information allocated to the sound source is limited, simple modeling of the sound source is performed. Therefore, there is a disadvantage that the reproduced sound is generally of low quality. Further, when the bit rate is reduced to 2400 bps or less, there is a disadvantage that the quality is further reduced because the analysis frame length needs to be increased. The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a speech encoding communication system having a high compression rate and a high naturalness.

[0005]

According to the speech coded communication system of the present invention, a discretized and framed line spectrum pair parameter of a frame unit of an input speech signal and a code book are previously obtained and stored by training. A code index of an optimal vector selected by pattern matching with a code vector, voiced / unvoiced information of a sound source signal in a frame unit obtained by removing the line spectrum pair parameter from the input voice signal, a pitch period of voiced section, and power information. the in the analysis synthesis system speech coding communication method for transmitting by multiplexing, dividing means for dividing a sound source signal of the frame into a plurality of subframes on the time axis, for each of said plurality of <br/> subframes A representative which judges voiced / unvoiced, extracts a pitch period of a voiced section, detects a representative pitch period from among them, and encodes it. A pitch period information output means, each subfolder
A plurality of pitch period difference information output means for quantizing the difference between the pitch period of the voiced section of the frame and the representative pitch period and encoding the difference together with voiced / unvoiced information; and calculating the power for each of the plurality of subframes. representative power information output means for encoding and detecting a representative power, a plurality of power difference information output means for quantizing a difference between the power and the representative power of each sub-frame, and the code index, each sub with voiced / unvoiced information of the frame, the representative pitch period information, the plurality of pitch period difference information, the representative power information, and multiplexing means for transmitting to a transmission path by multiplexing said plurality of power difference information, the frame Line specs for each
Code index of the optimal vector of
And for each frame excluding the line spectrum pair parameter
Representative pitch circumference of a subframe in which sound source information is divided into multiple parts
Period information, pitch period difference information of voiced sections, and voiced / unvoiced
Voice information, representative power information of the subframe, and each subframe.
And the power difference information of the
Transmission . Ma
Further , the speech coded transmission apparatus of the present invention includes a line spectrum pair parameter analyzer for extracting a line spectrum pair parameter from a discretized and framed input speech signal, and a line spectrum pair parameter which is learned by training in advance. A codebook stored as a vector, a vector quantizer that outputs a code index of an optimal vector selected by performing parameter matching between a line spectrum pair parameter extracted by the line spectrum pair parameter analyzer and a code vector of the codebook. And an inverse filter that calculates a sound source signal obtained by removing the line spectrum pair parameter from the input audio signal in frame units;
A sub-frame divider that divides the sound source signal of each frame into a plurality of sub-frames on a time axis, and determines voiced / unvoiced of each of the divided sub-frames and extracts a pitch period of a voiced section. a plurality of sub-frame pitch extractor, a representative pitch detector for determining a representative pitch period from the pitch period of the extracted <br/> issued said voiced interval of each sub-frame,
A representative pitch encoder for outputting a representative pitch cycle information surrogate table pitch period is coded, the said representative pitch cycle
A plurality of subframe pitch difference quantizers each of which quantizes a difference from a pitch period of a voiced section of each subframe and outputs encoded pitch period difference information together with voiced / unvoiced information; A plurality of sub-frame power extractors for calculating the power of each of the sub-frames; a representative power detector for determining the representative power from the calculated power of each sub-frame; and encoding the representative power to output representative power information. a representative power encoder, a plurality of sub-frame differential power quantizer for outputting a power difference information a difference between the representative power and the power of each sub-frame and each quantized, and the code index, the representative pitch cycle Information , said pitch period difference information and voiced / unvoiced information;
The representative power information, the power difference information and is the ash provided with a multiplexer to be transmitted to the transmission path by multiplexing.
Further, in the speech coded receiving device of the present invention, the code index of the optimal vector of the line spectrum pair parameter for each frame and the sound source information for each frame excluding the line spectrum pair parameter are divided into a plurality via the transmission path. Receiving a speech coded signal in which the representative pitch period information of the subframe, the pitch period difference information of the voiced section and the voiced / unvoiced information, and the representative power information of the subframe and the power difference information of each subframe are multiplexed; A speech coded receiving apparatus for decoding and outputting a reproduced speech, wherein the speech coded signal includes the code index, the representative pitch cycle information, pitch cycle difference information of a voiced section of each subframe, and voiced / unvoiced information, the representative power information, the <br/> a separator for separating output to the power difference information for each sub-frame, before separated The pitch period of the representative pitch and decoder, voiced segments of said each subframe decodes the pitch period difference information of each subframe, based on the surrogate table pitch period to output a representative pitch cycle decodes the representative pitch cycle information A plurality of subframe pitch decoders that output
A representative power decoder outputs a representative power by decoding the separated the representative power information, a plurality of output power of said each sub-frame by decoding the power difference information for each sub-frame based on the surrogate table power Subframe power decoder and previous
A plurality of pitch regenerator serial subframe reproduces the sound source signal from said decoded pitch period and power when voiced, noise the subframe is the sound source signal corresponding to the decoded power when silent An output noise generator, and the plurality of pitch reproducers or the noise generator according to voiced / unvoiced information in subframe units based on voiced / unvoiced information and pitch cycle difference information of each of the subframes output from the separator. A switch for switching the output, and a codebook searcher for reading out and outputting a line spectrum pair parameter corresponding to the code index output from the separator from the line spectrum pair parameter codebook having the same content as that of the transmission side; A voiced or unvoiced sound source signal output from the switcher and output from the codebook searcher. Provided with a synthesis filter for reproducing and outputting an audio signal and a line spectrum pair parameter is ash <br/> of.

That is, in the present invention, in order to solve the above-mentioned problem, voiced sound / unvoiced sound is finely determined by finely modeling the sound source of the analysis / synthesis type speech coding method on the time axis. In the case of vocal sounds, the gist is that the temporal change in pitch is finely expressed to improve the quality of reproduced voice and improve naturalness.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In the above-mentioned speech coded communication system of the present invention, first, spectrum parameters are extracted from an input speech signal discretized and framed by using spectrum analysis. Using the spectral parameters, the input speech signal is subjected to an inverse filter to calculate a prediction residual signal, that is, a source signal, and the source signal is divided into a plurality of subframes.
A voiced sound / unvoiced sound is determined for each of the divided subframes, and a pitch cycle for voiced sound is detected. A representative representative pitch is selected and encoded from a plurality of detected pitch periods, and the difference between the representative pitch and the detected pitch of each subframe is quantized and input to the multiplexer.

On the other hand, the power (power) of each of the divided subframes is calculated by the sum of squares, the largest one is detected from a plurality of detected powers, encoded as a reference power (representative power), and the representative power is selected. The difference between the power and the calculated power of each subframe is quantized and input to a multiplexer.

Further, the spectral parameters are obtained in advance by training, are subjected to pattern matching with code vectors stored in a code book, and an optimum vector is selected, and its index is input to a multiplexer. The above multiplexer input is multiplexed and transmitted to the transmission path.

In the decoding process of the multiplexed signal received via the transmission path, the received representative representative pitch is decoded, the pitch of each differentially quantized subframe is decoded, and the representative representative frame is decoded. The power is decoded, the differentially quantized subframe power is decoded, and the sound source signal is reproduced. Further, the code vector of the received index is read out from the code book as a spectral parameter, and the reproduced sound source signal is synthesized by a synthesis filter to obtain a reproduced sound output.

[0011]

3 and 4 are block diagrams showing an embodiment of the present invention. FIG. 3 shows a configuration on the transmission side, and FIG. 4 shows a configuration on the reception side. This will be described with reference to FIGS. In FIG. 3 on the transmitting side, 1 is an LSP parameter analyzer, 2 is an inverse filter, 3 is a subframe divider, 4 is a pitch extractor for each of the divided subframes, 5 is a representative pitch detector, 6 is a representative pitch detector. A pitch encoder, 7 is a pitch difference quantizer for each of the divided subframes, 8 is a power extractor for each of the divided subframes, 9 is a representative power detector, 10 is a representative power encoder, and 11 is a representative power encoder. A power difference quantizer 14 for each of the three divided subframes is a multiplexer.

In FIG. 4 on the receiving side, 15 is a separator, 1
6 is a representative pitch decoder, 17 is a switch, 18 is a pitch decoder for each of the divided subframes, 19 is a representative power decoder, 20 is a power decoder for each of the divided subframes, and 21 is 3 Reference numeral 22 denotes a noise generator, reference numeral 23 denotes a codebook searcher, reference numeral 12 denotes a codebook having the same contents as those on the transmission side, reference numeral 24 denotes a coefficient interpolator, and reference numeral 25 denotes a synthesis filter .

On the transmitting side shown in FIG. 3, the audio signal is first discretized and framed. The analysis frame length is about 45 ms to make it easy to reduce the bit rate. In the present invention, LSP (line spectrum pair) analysis was used for spectral parameter analysis. That is, the LSP parameter analyzer 1 converts the audio signal into LSPs in units of frames.
Retrieve a parameter. The inverse filter 2 calculates a sound source signal (prediction residual) by removing the LSP parameter obtained by the LSP parameter analyzer 1 from the input speech signal.
Next, the sound source signal is analyzed and modeled. In order to maintain naturalness, the prediction residual is divided into a plurality of subframes.
Voice / unvoice determination is performed in each subframe to detect a pitch period of a voiced section. Therefore, the sub-frame divider 3 divides the sound source signal into a plurality of sub-frames. The number of divisions is preferably 2 to 5 divisions, but in this embodiment, it is 3 divisions.

Next, a voiced / unvoiced determination and a pitch period are detected by three subframe pitch extractors 4 provided for each of the divided subframes. Conventionally, all detected pitch periods are coded, but this does not contribute to lowering the bit rate because the amount of information allocated to the sound source is greatly increased. Therefore, the voiced pitches obtained for each subframe are compared, one representative pitch is determined and encoded and transmitted, and the other pitches are obtained by encoding the difference between the quantized representative pitch and unvoiced information. Transmit. That is, the representative pitch is determined by the representative pitch detector 5 from the pitch cycle of each subframe obtained by the subframe pitch extractor 4. There are various methods for selecting the representative pitch, such as a maximum value, an intermediate value, a minimum value, and an average value. However, when the intermediate value and the average value are used as the representative values, it is necessary to consider the difference between both ±. In the present invention, the maximum value is set as the representative value. The representative pitch detected in this way is encoded by the representative pitch encoder 6.

On the other hand, the three sub-frame pitch difference quantizers 7 quantize the difference between the representative pitch coded by the representative pitch encoder 6 and the detected pitch period of each sub-frame, respectively, and perform voiced / unvoiced operations. Encode with information.

In quantizing the difference, the number of allocated bits must be considered in order to contribute to a lower bit rate. Since the difference between the subframes is small, the number of allocated bits is about 2 bits. Therefore, in the present embodiment, two bits are assigned to the difference amount and the voiced / unvoiced sound flag. For an unvoiced sound, the binary number is “00”, when the difference amount is 0 to −1, “01”, and the difference amount is −2. It is represented by 2 bits by setting “10” when the difference is −3 and “11” when the difference is -4 or more. By encoding the excitation signal for each sub-frame by this method, the fluctuation of the pitch with the passage of time is expressed without increasing the amount of information so much that the naturalness is improved.

On the other hand, the excitation signals of a plurality of (three) subframes divided by the subframe divider 3 are input to three subframe power calculators 8, respectively, and the power is calculated by the sum of squares of the subframe data. . As in the case of pitch, even when encoding power, a representative value is obtained and encoded in advance, and the difference is quantized. That is, the representative power is determined by the representative power detector 9 from the power information for each subframe obtained by the three subframe power calculators 8. The determined representative power is encoded by the representative power encoder 10. The three subframe power difference quantizers 11 quantize the difference between the representative power encoded by the representative power encoder 10 and the detected power of each subframe. In quantizing the difference amount, the number of allocated bits is set to 2 bits, as in the case of the pitch, since the difference amount between the subframes is small.

Next, vector quantization of the LSP parameters obtained by the LSP parameter analyzer 1 is performed. In vector quantization, LSP parameters are learned in advance by a training signal, and are stored in the LSP parameter codebook 12 as code vectors. The vector quantizer 13 calculates the L obtained by the LSP parameter analyzer 1
SP parameter and LSP parameter code book 12
, And an optimal vector is selected to obtain a code index.

The multiplexer 14 includes a code index from the vector quantizer 13, representative pitch information encoded by the representative pitch encoder 6, and each subframe encoded by the subframe pitch difference quantizer 7. Pitch information and voiced / unvoiced information, the representative power information encoded by the representative power encoder 10, and the subframe power difference quantizer 1
1 is multiplexed with the power information of each subframe coded by 1 and output to the transmission path.

Next, the receiving side of FIG. 4 will be described. The multiplexed signal received via the transmission path is input to the demultiplexer 15 and coded representative pitch information, coded subframe pitch difference pitch cycle and voiced / unvoiced information of each subframe, and The obtained representative power information, the subframe power difference of each coded subframe, and the code index are output separately.

The representative pitch decoder 16 decodes the encoded representative pitch information. The switch 17 outputs the output of the three pitch reproducers 21 or the output of the noise generator 22 according to the voiced / unvoiced information for each subframe from the differentially quantized subframe pitch information which is also differentially quantized and also serves as voiced / unvoiced information. And output. When voiced, the subframe pitch decoder 18 decodes the differentially quantized subframe pitch information based on the representative pitch.

The representative power decoder 19 decodes the encoded representative power information. The subframe power decoder 20 decodes the differentially quantized subframe power information based on the decoded representative power. In the case of voiced sound, in three cases, the pitch reproducer 21 reproduces the sound source signal from the subframe pitch information and the subframe power information, and in the case of unvoiced sound, the noise generator 22 reproduces the sound source signal from the subframe power information. .

The codebook search unit 23 refers to the code in the codebook 12 and sets the code index to the LSP.
Replace with the corresponding LSP parameter in the parameter codebook. The synthesis filter 25 includes a voiced or unvoiced sound source signal output from the switch 17 and a coefficient-interpolated L
The audio signal is reproduced and output from the SP parameter. 45ms
Because the spectral envelope is considered to be distorted by the analysis frame having a length as long as possible, the LSP parameter interpolator 24 allows the analysis to be performed over the first half of the analysis frame.
Degradation is suppressed by performing linear interpolation between the LSP parameter of the past frame and the LSP parameter of the current frame.

[0024]

As described above, according to the present invention, in a voice coded communication for converting a voice signal into a digital signal format and transmitting the digital signal, modeling of a sound source of an analysis-synthesis type voice coding system is performed. By making finer on the time axis, the time change of the pitch of the voice can be expressed, and the voiced / unvoiced sound can also be finely determined, so that the deterioration of the received and reproduced voice quality at low bit rates is reduced and the naturalness is improved. Therefore, the practical effect is great.

[Brief description of the drawings]

FIG. 1 is a conceptual diagram of a conventional CELP.

FIG. 2 is a block diagram of a conventional LPC.

FIG. 3 is a block diagram showing an embodiment (transmission side) of the present invention.

FIG. 4 is a block diagram showing an embodiment (receiving side) of the present invention.

[Explanation of symbols]

1 LSP parameter analyzer 2 Inverse filter 3 Subframe divider 4 Subframe pitch extractor 5 Representative pitch detector 6 Representative pitch encoder 7 Subframe pitch difference quantizer 8 subframe power extractor 9 Representative power detector 10 Representative power encoder 11 Subframe power difference quantizer 12 Codebook 13 Vector Quantizer 14 Multiplexer 15 Separator 16 representative pitch decoder 17 Switch 18 Subframe pitch decoder 19 Representative power decoder 20 subframe power decoder 21 Pitch regenerator 22 Noise generator 23 Codebook Searcher 24 coefficient interpolator 25 Synthesis filter

Claims (3)

(57) [Claims] 1. A code of an optimal vector selected by performing pattern matching on a line spectrum pair parameter in a frame unit of a discretized and framed input speech signal and a code vector previously obtained and stored in a code book by training. An index, voiced / unvoiced information of a sound source signal in frame units obtained by removing the line spectrum pair parameter from the input voice signal, a pitch period of a voiced section,
In analysis-synthesis system speech coding communication method for transmitting by multiplexing and power information, and dividing means for dividing into a plurality of sub-frame sound source signal of the frame on the time axis, voiced each said plurality of sub-frame A representative pitch cycle information output means for judging / unvoiced, extracting a pitch cycle of a voiced section, detecting and encoding a representative pitch cycle therefrom, a pitch cycle of a voiced section of each subframe and the representative pitch cycle A plurality of pitch period difference information output means for quantizing the difference between the subframe and the voiced / unvoiced information, and representative power information for calculating power for each of the plurality of subframes, detecting representative power from among them, and coding and output means, a plurality of power difference information output means the quantizing a difference between power and the representative power of each sub-frame, and the code index, each sub Voiced / unvoiced information frame, the representative pitch period information, the plurality of pitch <br/> Chi period difference information, the representative power information, the plurality of power difference information and multiplexing means for transmitting to a transmission path by multiplexing And the optimal vector of the line spectrum versus parameters for each frame
And the line index versus the parameter
Sound source information for each frame excluding data
Lame representative pitch period information and pitch period of voiced section
Difference information, voiced / unvoiced information, and the representative
Power information and power difference information of each subframe are multiplexed.
A voice coded communication system characterized by transmitting a voice coded signal . 2. A line spectrum pair parameter analyzer for extracting a line spectrum pair parameter from a discretized and framed input speech signal, and a codebook in which a line spectrum pair parameter is previously learned by training and stored as a code vector. A vector quantizer that outputs a code index of an optimal vector selected by performing parameter matching between the line spectrum pair parameter extracted by the line spectrum pair parameter analyzer and the code vector of the codebook, and from the input audio signal. and inverse filter to calculate the excitation signal except for the line spectral pair parameters for each frame, and the sub-frame divider for dividing into a plurality of sub-frame sound source signal of the frame on the time axis, each of the sub, which are the divided Judgment of voiced / unvoiced frame and voiced A plurality of sub-frame pitch extractor for extracting a pitch period of the segment, the representative pitch detector for determining a representative pitch period from the pitch period of the voiced of each subframe the extraction, the surrogate table pitch period A representative pitch encoder that encodes and outputs representative pitch cycle information; and a pitch cycle difference that quantizes a difference between the representative pitch cycle and a pitch cycle of a voiced section of each subframe and encodes the difference together with voiced / unvoiced information. A plurality of subframe pitch difference quantizers that output information; a plurality of subframe power extractors that calculate the power of each of the divided subframes; and a representative power among the calculated powers of the subframes. a representative power detector for determining a representative power encoder that outputs a representative power information by encoding the surrogate table power, the representative power and the respective sub A plurality of sub-frame power difference quantizers each outputting a power difference information obtained by quantizing a difference between the power of the frame and the power of the frame; the code index, the representative pitch cycle information, the pitch cycle difference information, and voiced / unvoiced information. And a multiplexer for multiplexing the representative power information and the power difference information and transmitting the multiplexed power information to a transmission path. 3. A code index of an optimal vector of a line spectrum pair parameter for each frame via a transmission path, and representative pitch period information of a subframe obtained by dividing a plurality of sound source information for each frame excluding the line spectrum pair parameter. And a voice encoded signal in which the pitch cycle difference information and voiced / unvoiced information of the voiced section, the representative power information of the subframe, and the power difference information of each subframe are multiplexed and decoded to output a reproduced voice. A speech coded receiving device, comprising: encoding the speech coded signal with the code index, the representative pitch cycle information , pitch cycle difference information of a voiced section of each subframe, and voiced / unvoiced information ; and br /> table power information, wherein the separator for separating output to the power difference information for each sub-frame, a separated the representative pitch period information A representative pitch decoder for outputting a representative pitch cycle in issue, a plurality of outputs the pitch period of the voiced interval of the each sub-frame by decoding the pitch period difference information of each subframe, based on the surrogate table pitch period and the sub-frame pitch decoder, a representative power decoder decodes the separated the representative power information and outputs the representative power, each sub-frame by decoding the power difference information for each sub-frame based on the surrogate table power a plurality of sub-frame power decoder for outputting a power, a plurality of pitch regenerator said subframe to reproduce the sound source signal from said decoded pitch period and power when voiced, when the sub-frame is unvoiced A noise generator that outputs noise serving as a sound source signal corresponding to the decoded power;
A switch for switching and outputting the outputs of the plurality of pitch reproducers or the noise generator in accordance with voiced / unvoiced information in subframe units based on unvoiced information and pitch cycle difference information, and corresponding to the code index output from the separator A line spectrum pair parameter to be read out from the line spectrum pair parameter codebook having the same contents as the transmitting side and output, and a voiced or unvoiced sound source signal output from the switcher and output from the codebook searcher. And a synthesis filter for reproducing and outputting an audio signal from the line spectrum pair parameter to be output.