JPH10308708A - Voice encoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reproduce a signal without having feeling of interruption in speech even when an error that is not correctable is continuously detected in a voice encoder which is used on a transmission path that has many bit errors. SOLUTION: An error correcting and error detecting device 103 performs error correction and error detection of receiving data on a decoding side. When detection information does not detect an error in the section, a voice decoder 104 performs voice decoding, and an adder 106 outputs an output signal from the decoder 104 as decoded voice as it is. On the other hand, in the case of a section where errors are continuously detected, a comfortable noise generator 105 generates a comfortable noise signal by partially or entirely using a voice coding parameter including an error and makes an output signal of an error section by outputting it after an adder 106 adds an error section compensation signal that is outputted from the decoder 104 to a comfortable noise signal that is outputted from the generator 105.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding apparatus used for a digital mobile communication terminal such as a digital portable telephone.

[0002]

2. Description of the Related Art Conventionally, as a speech coding apparatus having an error correction function, for example, Channel Coding For Digital
Speech Transmission In Japanese Digital Cellular S
The one described in ystem (by MJMcLaughlin, The Institute of Electronics, Information and Communication Engineers, Radio Communication Systems Study Group, RCS90-27) is known. FIG. 9 shows the configuration of a conventional speech coding apparatus. On the encoding side (a), reference numeral 901 denotes a speech encoder, which performs speech encoding on input speech and outputs encoding parameters. An error correction encoder 902 performs error correction encoding and error detection code calculation on the encoding parameters. On the decoding side (b), 90
An error correction / error detector 3 performs error correction and error detection on received data. A speech parameter interpolator 904 interpolates speech encoding parameters at the time of error detection based on the speech encoding parameters after error correction and the error detection information. A speech decoder 905 performs speech decoding from speech encoding parameters.

[0003]

However, in the above conventional speech coding apparatus, there are many bit errors in the transmission path,
When errors that cannot be completely corrected by error correction occur continuously, the interpolation of the speech coding parameters by the speech parameter interpolator is continued and the output signal is processed so as to gradually mute the output signal. In a wireless transmission line having a low transmission speed, there is a problem that an output signal is interrupted over a long section, resulting in deterioration in audibility.

[0004] The present invention solves the above-mentioned conventional problem, and generates comfort noise using a part of speech coding parameters received in a continuous error detection section, and reproduces decoded speech without a sense of interruption. It is therefore an object of the present invention to provide an excellent speech coding apparatus capable of suppressing audible deterioration and reproducing received speech information with low quality even during an error section.

[0005]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problem, the present invention provides a method for detecting noise in a continuous error detection section by using a part of speech coding parameters received in the section, in particular, spectrum parameters. Is generated, and the comfort noise signal is output instead of the decoded speech signal. In addition, in addition to the configuration with a normal speech encoder and an error correction encoder, a speech encoder with high error tolerance is provided in parallel, and the decoding side is configured for the encoding parameters of the normal speech encoder. When an error is detected, the output of a speech decoder with high error resilience provided in parallel is used as a decoded speech output in the error detection section.

As described above, even in a continuous error detection section, by outputting a comfort noise signal including information of a received parameter or a decoded speech signal by a speech encoder having high error resilience, decoded speech is reproduced without interruption. In addition to suppressing audible deterioration, it is possible to reproduce received audio information with low quality even during an error section.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The invention according to claim 1 of the present invention provides, on the encoding side, a speech coder for performing speech coding on input speech and outputting speech coding parameters. An error correction encoder that performs error correction coding and error detection code calculation on the coding parameter and outputs transmission data, and provides an error correction / error detection / error detection for received data on the decoding side. An error detector, a speech decoder that performs speech decoding from the speech coding parameters after error correction, and a comfort noise generator that generates comfort noise using a part of the speech coding parameters for the error detection interval. In the section where error detection is continuous, by generating comfort noise using a part of the received speech coding parameters of the error detection section and outputting it as an output signal of the section, Even during the connection error detection section,
There is an effect that the received audio information can be reproduced with low quality, and a sense of discontinuity, which is a factor of auditory deterioration, can be eliminated.

[0008] The invention described in claim 2 of the present invention provides:
In the invention according to claim 1, the encoding parameter includes:
In a configuration including at least spectral information such as LPC parameters representing vocal tract information of input speech and sound source information corresponding to vocal cord information, the comfort noise generator on the decoder side uses the spectral information among the received speech coding parameters. A speech spectrum decoding unit for decoding speech spectrum parameters, a noise source generation unit for generating a noise source, and a synthesis filter for synthesizing and outputting a comfort noise signal. By composing a synthesis filter using all or a part of the spectrum information and generating a comfort noise using it and outputting it as an output signal of the section, even during the continuous error detection section,
There is an effect that the received phoneme information can be reproduced while having low quality, and a sense of discontinuity which is a factor of auditory deterioration can be eliminated.

Further, the invention according to claim 3 of the present invention provides:
In the invention according to claim 2, the quantizer of the speech spectrum parameter in the speech encoder on the encoding side is 1
A multi-stage configuration having a scalar quantization unit for each order of the spectrum parameter at the stage, and a divided vector quantization unit for dividing the spectrum parameter into a plurality of orders and vector-quantizing each vector after the second stage For the error detection section on the decoding side, the spectrum decoding unit in the comfort noise generator decodes the spectrum parameters using only the first-stage scalar quantization code among the received spectrum parameter codes. By doing so, there is an effect that the spectrum can be decoded with less influence of errors while having low quality.

[0010] The invention described in claim 4 of the present invention provides:
In the invention according to claim 2 or 3, the spectrum decoding unit in the comfort noise generator on the decoding side decodes a spectrum parameter from a received spectrum parameter code of the section, and a spectrum parameter representing a white noise spectrum. A white noise spectrum buffer that stores a decoded spectrum parameter and a synthesized spectrum calculation unit that generates a spectrum parameter for synthesizing a comfort noise signal from the white noise spectrum parameter. Has an effect that more natural error section signal reproduction can be performed in the error section by operating such that it approaches the white noise spectrum.

Further, the invention according to claim 5 of the present invention provides:
In the invention according to claim 2 or 3, the spectrum decoding section in the comfort noise generator on the decoding side decodes a spectrum parameter from a received spectrum parameter code of the section and a spectrum of a past silent section. A silence section spectrum buffer that stores spectrum parameters to be represented, and a synthesized spectrum calculation section that generates a spectrum parameter for synthesizing a comfortable noise signal from the decoded spectrum parameter and the silence section spectrum parameter, and where error detection is continuous By operating the synthesized spectrum asymptotically to the spectrum of the previous silent section, a more natural error section signal reproduction having a spectral characteristic similar to the reproduced signal of the past silent section in the error section is performed. Has the effect of being able to

[0012] The invention described in claim 6 of the present invention provides:
In the invention according to any one of claims 2 to 5, the noise source generator in the comfort noise generator on the decoding side stores a silent section source signal buffer for storing a source signal in a silent section, and a noise source using the same. And a noise sound source generation unit that generates
By using the sound source signal of the past silent section as the sound source signal of the error detection section, it is possible to perform more natural error section signal reproduction having a sound source characteristic similar to the reproduction signal of the past silent section in the error section. Has an action.

The invention according to claim 7 of the present invention provides:
On the encoding side, a high-efficiency speech encoder that performs high-efficiency speech encoding at a low bit rate such as CELP encoding on input speech, and error correction encoding for the encoding parameters obtained thereby. And an error correction encoder for calculating an error detection code, and a high error resilience encoder capable of realizing high error resilience with low quality such as ADM coding for input speech. Among them, an error correction / error detector that performs error correction and error detection on the received data corresponding to the output of the high-efficiency encoder, and speech decoding corresponding to the high-efficiency encoder based on the speech coding parameters after error correction A high-efficiency speech decoder for performing speech decoding, a high-error-tolerant speech decoder for performing speech decoding on received data corresponding to the output of the high-error-tolerant speech encoder among the received data, and Are those with an adder which outputs as an output signal of the section by switching or adding outputs from efficient voice decoder and high error resilience speech decoder,
When the error detection is continuous, the output from the high-error-tolerant speech decoder is used as a decoded speech signal, so that the decoded speech can be reproduced without interruption.

Further, the invention according to claim 8 of the present invention provides:
8. The encoding apparatus according to claim 7, further comprising a subtracter on the encoding side for subtracting a decoded speech signal of the high error resilience encoding by the high error resilience encoder from the input speech. By performing high-efficiency speech encoding by the encoder on the signal subtracted from the input speech by the decoded speech signal of the high error resilience encoding performed by the high error resilience encoder performed earlier, error detection can be performed continuously. In this case, there is an effect that it is possible to achieve both reproduction of a decoded sound without a sense of interruption of the sound and reduction of the bit rate.

The invention according to claim 9 of the present invention provides:
A recording medium such as a magnetic disk, a magneto-optical disk, a ROM cartridge, or the like, on which a program that implements the audio encoding device according to claim 1 by software is recorded, or an audio encoder using the recording medium. This is a device that operates, and has an effect that the speech encoding device of the present invention can be realized by software.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. (Embodiment 1) FIG. 1 is a block diagram of a speech coding apparatus according to a first aspect of the present invention. On the code side (a), 101 is a speech coder that performs speech coding on input speech and outputs speech coding parameters, and 102 is error correction coding and error detection on the obtained coding parameters. An error correction encoder that performs code calculation and outputs transmission data. On the decoding side (b), 103 is an error correction / error detector that performs error correction and error detection on received data, 104 is a speech decoder that performs speech decoding from speech coding parameters after error correction, and 105 is an error decoder. A comfort noise generator that generates comfort noise using a part of speech coding parameters for a detection section,
An adder 106 adds the output from the speech decoder 104 and the output from the comfort noise generator 105 according to the error detection information.

The operation of the speech coding apparatus configured as described above will be described with reference to FIG. First, on the encoding side (a), speech encoding is performed by the speech encoder 101 on an input speech signal segmented into fixed short sections, and speech encoding parameters are output. Then, an error correction encoder 102
Performs error correction coding and error detection code calculation, and outputs the result as transmission data. In many cases, the error correction encoder 102 calculates an error correction code and an error detection code only for parameters that are perceptually important among speech coding parameters. Next, on the decoding side (b), the error correction / error detector 103 performs error correction and error detection on the received data. The error detection is to detect whether an error remains in the received data after the error correction, and a CRC code is often used. If the target of error detection is limited to auditory important parameters,
It is detected whether an error remains in the error detection target parameter. If no error is detected by the error detection information in the section, the speech decoder 104 performs speech decoding, and the adder 106 outputs the output signal from the speech decoder 104 as it is as a decoded speech. On the other hand, when the error is detected in the section, the speech decoder 104 first performs error section compensation processing. This generates a compensation signal for an error section using the past coding parameters or decoding drive excitation / decoded speech. However,
If error sections continue, a process of gradually muting is added. Next, in the comfort noise generator 105, a comfort noise signal is generated using a part or all of the speech coding parameters including the error. This is generated so that a part of the audio information (for example, spectrum information, power information, etc.) obtained from the reception data in the error section is reflected in the comfort noise to be output. The generation method is arbitrary. Then, the error section compensation signal output from the speech decoder 104 and the comfort noise signal output from the comfort noise generator 105 are weighted and added by the adder 106, and output as error section output signals.

As described above, according to the first embodiment of the present invention, in a section where error detection is continuous, comfort noise is generated using a part of the received speech coding parameters in the error detection section. By outputting the output signal as an output signal of the section, the received voice information can be reproduced with low quality even during the continuous error detection section, and the sense of discontinuity, which is a cause of auditory deterioration, can be eliminated.

(Embodiment 2) FIG. 2 shows a second embodiment of the present invention.
FIG. 9 is a block diagram of the comfort noise generator 105 on the decoding side of the speech coding apparatus according to the first embodiment. In FIG. 2, reference numeral 201 denotes a speech parameter separation unit that separates received speech coding parameters into respective speech coding parameters; 202, a spectrum decoding unit that decodes speech spectrum parameters using spectrum information among the received speech coding parameters; Is a noise source generation unit that generates a noise source, 204 is a synthesis filter that synthesizes and outputs a comfort noise signal, and 205 is a level control unit that controls the level of the synthesized signal by the synthesis filter 204.

The comfort noise generator 1 configured as described above
The operation of the device 05 will be described with reference to FIG. The second embodiment can be applied to voice coding including at least spectral information such as LPC parameters representing vocal tract information of input voice and sound source information corresponding to vocal cord information in coding parameters. First, the speech parameter separating unit 201 obtains a spectrum parameter representing spectrum information from the received speech coding parameters. Then, spectrum decoding section 202 decodes the speech spectrum from the received spectrum parameters. In general, LPC parameters (for example, LPC coefficients and LSP parameters) are often used as voice spectrum parameters. Next, the noise sound source generation unit 203 generates a driving sound source signal of comfort noise. A stationary noise signal such as white random noise is used as the driving sound source signal. Then, the synthesis filter 204 synthesizes the sound source signal obtained by the noise sound source generation unit 203 and the comfort noise signal using the spectrum parameter obtained by the spectrum decoding unit 202, and controls the output level by the level control unit 205. And then output.

As described above, according to the second embodiment of the present invention, in a continuous error detection section, a synthesis filter is configured by using all or a part of the received speech spectrum information of the error section, and a comfortable filter is formed by using the synthesis filter. By generating noise and outputting it as an output signal in that section, even during a continuous error detection section, received phonological information can be reproduced with low quality, and a sense of discontinuity, which is a cause of auditory deterioration, can be eliminated. .

(Embodiment 3) FIG. 3 shows a third embodiment of the present invention.
In the case of a configuration in which the speech encoder on the encoding side of the speech encoding apparatus according to Embodiment 1 includes a spectrum quantizer for quantizing spectrum information representing vocal tract information of speech, FIG. 10 is a block diagram of a spectrum decoding unit 202A in a comfort noise generator 105 on the decoding side according to Embodiment 2 and a decoding unit 107 in Embodiment 2. In FIG. 3A, reference numeral 301 denotes a scalar quantization unit for each order of a spectrum parameter in a voice spectrum parameter quantizer, and 302 denotes a division for dividing a spectrum parameter into a plurality of orders and performing vector quantization for each vector. This is a vector quantization unit. In FIG. 3B, reference numeral 303 denotes a scalar quantization decoding unit that decodes a scalar quantization part in a spectrum decoding unit 202A of the comfort noise generator 105, and 304 denotes a division vector quantization that decodes a division vector quantization part. It is a decoding unit.

The operation of the speech coding apparatus configured as described above will be described with reference to FIG. First, in the spectrum quantizer 107 on the encoding side, a scalar quantization unit 301 of the first stage is applied to an input spectrum (vector).
Scalar-quantizes the scalar value for each order. Next, the division vector quantization unit 302 quantizes the first-stage quantization error portion as a second-stage quantization unit. Here, the spectral parameter is divided into a plurality of orders, and quantization is performed by vector quantization with high quantization efficiency. Then, the quantization codes (scalar quantization codes,
(A division vector quantization code). On the other hand, also in the audio decoder on the decoding side, a quantization decoder is configured with the same configuration as described above to perform spectrum decoding. That is, spectrum decoding section 202 in comfort noise generator 105 on the decoding side.
In A, the spectrum of the scalar quantization part is decoded by the scalar quantization decoding unit 303 using only the scalar quantization code among the received spectrum codes, and is output as a spectrum of comfort noise.

As described above, according to the third embodiment of the present invention, the spectrum quantization on the encoding side has a multi-stage configuration, and the scalar quantization is less affected by bit errors at the first stage although the quantization efficiency is reduced. Is configured to perform vector quantization with high quantization efficiency in the second stage, and the spectrum decoding unit 2 of the comfortable noise generator 105
In 02A, by using the decoded spectrum of only the scalar quantized portion of the first stage having high error resilience as a synthesized spectrum of comfort noise, it is possible to perform spectrum decoding with low quality and less influence of errors.

In the third embodiment, a two-stage scalar quantization + vector quantization is shown.
A multi-stage configuration of more than stages may be used. In addition, the scalar quantization unit 301 of the first stage uses the difference from the quantization result of the first stage of the previous section or predictive quantization, so that the number of scalar quantization bits is 1 and the quantization efficiency is improved. Thus, decoding with less influence of errors can be performed. It is also possible to use a quantizer having a quantization structure with high error tolerance.

(Embodiment 4) FIG. 4 shows a fourth embodiment of the present invention.
FIG. 13 is a block diagram of a spectrum decoding unit 202B in the comfort noise generator 105 on the decoding side of the speech encoding apparatus according to claim 2 or 3. In FIG. 4, reference numeral 401 denotes a section spectrum decoding unit for decoding a spectrum parameter from a received spectrum parameter code of the section, 402 denotes a white noise spectrum buffer storing a spectrum parameter representing a white noise spectrum, 403 denotes a decoded spectrum parameter and This is a combined spectrum calculation unit that generates a spectrum parameter for combining a comfort noise signal from a white noise spectrum parameter.

The operation of spectrum decoding section 202B configured as described above will be described with reference to FIG. First, the section spectrum decoding section 401 performs spectrum decoding from the received spectrum parameter code of the section. This may be either decoding using all received spectrum codes or decoding using only partial codes shown in the third embodiment. Next, the combined spectrum calculation section 403 calculates a combined spectrum parameter used in the section using the spectrum parameter output from the section spectrum decoding section 401 and the white noise spectrum parameter output from the white noise spectrum buffer 402. The composite spectrum parameter is calculated such that when the error section is continuous over a long section, a spectrum is obtained such that the composite spectrum gradually approaches the white noise spectrum. For example, as a spectral parameter, L
When the SP parameter is used, linear addition of the section decoding LSP and the white noise LSP is performed, and the weighting coefficient is controlled so as to approach the white noise LSP in proportion to the number of continuous sections for error detection. Other implementations are possible.

As described above, according to the fourth embodiment of the present invention, when error detection continues for a long period of time, the composite spectrum operates so as to approach white noise, so that the error section , More natural error section signal reproduction can be performed.

(Embodiment 5) FIG. 5 shows a fifth embodiment of the present invention.
FIG. 10 is a block diagram of a spectrum decoding unit 202C in the comfort noise generator 105 on the decoding side of the speech encoding apparatus according to claim 2 or 3. In FIG. 5, reference numeral 501 denotes a section spectrum decoding unit for decoding a spectrum parameter from a received spectrum parameter code of the section, and 502 stores a spectrum parameter representing a spectrum of a past silent section (including a section including only noise). Silent section spectrum parameter buffer, 50
Reference numeral 3 denotes a synthesized spectrum calculation unit that generates a spectrum parameter for synthesizing a comfortable noise signal from the decoded spectrum parameter and the silent section spectrum parameter.

The operation of spectrum decoding section 202C configured as described above will be described with reference to FIG. First, the section spectrum decoding section 501 performs spectrum decoding from the received spectrum parameter code of the section. This may be either decoding using all received spectrum codes or decoding using only partial codes shown in the third embodiment. On the other hand, the silence section spectrum parameter buffer 502 receives the decoded spectrum parameters of the section determined as the past silence (including the section with no voice and the section with only the ambient noise) and sequentially averages them. , And holds an average spectrum parameter of a silent section. Next, the synthesized spectrum calculation unit 503 calculates a synthesized spectrum parameter used in the section using the spectrum parameter output from the section spectrum decoding unit 501 and the silent section spectrum parameter output from the silent section spectrum parameter buffer 502. The composite spectrum parameter is calculated such that when the error section is continuous, a spectrum is obtained such that the composite spectrum gradually approaches the silent section spectrum. For example, when the LSP parameter is used as the spectrum parameter, linear addition of the section decoding LSP and the silent section LSP is performed, and the weighting coefficient is controlled so as to approach the silent section LSP in proportion to the number of continuous sections for error detection. Other implementations are possible.

As described above, according to the fifth embodiment of the present invention, when error detection is continued for a long period of time, the combined spectrum operates so as to approach the spectrum in a silent section, thereby reducing the error. It is possible to perform more natural error section signal reproduction having a spectrum characteristic similar to the reproduction signal of the past silent section in the section.

(Embodiment 6) FIG. 6 shows a sixth embodiment of the present invention.
FIG. 10 is a block diagram of a noise source generator 203 in a comfortable noise generator 105 on the decoding side in the speech encoding apparatus according to any one of claims 2 to 5. In FIG. 6, reference numeral 601 denotes a sound source signal buffer for storing a sound source signal in a silent period, and reference numeral 602 denotes a noise source generating unit for generating a noise source using the sound source signal in the silent period.

The noise source generator 2 configured as described above
The operation of No. 03 will be described with reference to FIG. First, the silent section sound source signal buffer 601 sequentially converts a sound source signal representative of a silent section from a driving sound source signal at the time of speech decoding of a section determined to be silent in the past (including a section without sound and a section including only ambient noise). That is simply stored in the buffer, or that the driving sound source signal of the previous silent section is simply added sequentially, or that the past driving sound source is correlated with the past stored buffer. A sound source signal having a sound source characteristic of a silent section is held by, for example, performing weighted addition of a section having a high sound field and updating. Then, a noise source signal is generated by randomly extracting a fixed section length from the silent section source signal buffer 601 by the noise source generating section 602.

As described above, according to the sixth embodiment of the present invention, by using the sound source signal of the past silent section as the sound source signal of the error detecting section, the error section is similar to the reproduced signal of the past silent section. It is possible to reproduce a more natural error section signal having the improved sound source characteristics.

(Embodiment 7) FIG. 7 shows a seventh embodiment of the present invention.
1 is a block diagram of a speech encoding device corresponding to FIG. In FIG. 7, on the encoding side (a), 701
Is a high-efficiency speech coder that performs high-efficiency speech coding at a low bit rate on input speech, and 702 is an error-correcting code that performs error correction coding and error detection code calculation on the obtained coding parameters. 703, a high error resilience encoder capable of realizing high error resilience with low quality to the input voice, and 704, a multiplex for multiplexing output data of the error correction encoder 702 and the high error resilience encoder 703. It is a gasifier. On the decoding side (b), reference numeral 705 denotes a separator for separating received data on the decoding side, and reference numeral 706 denotes an error correction / error correction unit for performing error correction and error detection on the received data corresponding to the output of the high-efficiency encoder out of the received data. An error detector 707 is a high-efficiency speech decoder that performs speech decoding corresponding to the high-efficiency speech coder 701 based on the speech coding parameters after error correction, and 708 is a high error-resistant speech coder 703 of the received data. A high error-tolerant speech decoder 709 that performs speech decoding on the received data corresponding to the output, and switches or adds outputs from the high-efficiency speech decoder 707 and the high error-tolerant speech decoder 708 according to the result of error detection. Is an adder that outputs as an output signal of the corresponding section.

The operation of the speech coding apparatus configured as described above will be described with reference to FIG. First, on the encoding side (a), a high-efficiency speech encoder 701 performs high-efficiency speech encoding at a low bit rate on input speech. CELP coding is a typical example of high-efficiency voice coding at a low bit rate. The error correction encoder 702 performs error correction encoding and error detection code calculation on the encoding parameters obtained by the high-efficiency speech encoder 701, and outputs encoded data. On the other hand, in parallel with this, the high error resilience encoder 703 performs high error resilience voice coding that can realize high error resilience with low quality for the input voice, and outputs encoded data without error correction coding. I do. As the speech coding with high error tolerance, ADM coding or a sample value of an input signal is + 1 / -1.
And quantization. In any case, the quality of the decoded signal is low at a low bit rate, but the error resilience to a bit error is high. Furthermore, by performing band limitation on the input voice and applying the band limitation only to the low frequency band, encoding can be performed at a lower bit rate. Then, the output data of the error correction encoder 702 and the high error resilience encoder 703 are multiplexed by the multiplexer 704,
Output as transmission data. Next, on the decoding side (b), the received data corresponding to the output of the high-efficiency speech encoder 701 and the high-error-resistant speech encoder 703 are output by the separator 705.
Separate the received data corresponding to the output. The error correction and error detector 706 performs error correction and error detection on the received data corresponding to the output of the high-efficiency encoder, and the high-efficiency audio decoder 707 extracts the error-corrected speech encoding parameters from the speech encoding parameters. Speech decoding is performed corresponding to the high-efficiency speech encoder 701. In addition, speech decoding is performed on the received data corresponding to the output of the high error tolerant speech encoder 703 among the received data by the high error tolerant speech decoder 708. Then, the output from the high-efficiency speech decoder 701 and the high-error-tolerant speech decoder 703 is switched or added according to the result of the error detection in the adder 709, and is output as an output signal of the section. Specifically, in the section where no error is detected, the output of the high-efficiency speech decoder 707 is output as an output signal as it is, and in the error detection section, the decoded signal by the error section compensation in the high-efficiency speech decoder 707 and the high error An added signal with the output signal of the robust speech decoder 708 is output, and when error detection continues,
The output signal from the high-error-tolerant speech decoder 708 is output.

As described above, according to the seventh embodiment of the present invention, when error detection is continuous, the output from high error-tolerant speech decoder 708 is a decoded speech signal.
The decoded audio can be reproduced without interruption.

(Embodiment 8) FIG. 8 shows an eighth embodiment of the present invention.
FIG. 10 is a block diagram showing a coding side of the speech coding apparatus according to claim 7. In FIG. 8, 80
Reference numeral 1 denotes a subtractor for subtracting the decoded speech signal of the high-error-tolerant speech encoding from the input speech, and the other configuration is the same as that of the speech encoding apparatus according to the seventh embodiment shown in FIG.

The operation of the speech coding apparatus configured as described above will be described with reference to FIG. First, on the encoding side, a high error-tolerant speech encoder 703 performs high-error-tolerant speech encoding that can realize high-error tolerance with low quality for input speech, and encodes encoded data without error correction coding. In addition to the output, the audio decoding is performed using the encoded data encoded by the high-error-tolerant audio main encoder 703, and a decoded signal is output. And the subtractor 801
Subtracts the decoded speech signal of the high error resilience speech coding by the high error resilience encoder 703 from the input speech and outputs the difference signal. Next, the high-efficiency speech encoder 701 performs high-efficiency speech encoding at a low bit rate using the difference signal as an input. And an error correction encoder 7
02, error correction coding and error detection code calculation are performed on the coding parameters obtained by the high-efficiency speech coder 701, and code data is output. Then, the multiplexer 704 multiplexes the output data of the error correction encoder 702 and the high error resistant speech encoder 703 and outputs the result as transmission data. On the other hand, the decoding side is the same as the decoding side of the seventh embodiment shown in FIG. However, in the adder 709, in the section where no error is detected, the output signal of the high-efficiency speech decoder 707 and the high-error-tolerant speech decoder 70 are used.
8 and outputs the added signals. If error detection continues, only the output signal from the high-error-tolerant speech decoder 708 is output.

As described above, according to the eighth embodiment of the present invention, high-efficiency speech encoding by high-efficiency speech encoder 701 is performed prior to high-error-resistant speech encoder 7.
By applying the decoded speech signal of the high error resilience coding according to C.03 to the signal subtracted from the input speech, it is possible to achieve both the reproduction of the decoded speech without any interruption of the speech and the lower bit rate when the error detection is continuous. Can be done.

(Embodiment 9) A ninth embodiment of the present invention is a magnetic disk, a magneto-optical disk, and a ROM cartridge in which a program that implements the audio encoding device according to any one of claims 1 to 8 by software is recorded. Etc., or a device that operates as a speech encoder using the recording medium, and the speech encoding device of the present invention can be realized by software.

[0042]

As described above, according to the first aspect of the present invention, in a section where error detection is continuous, comfort noise is generated using a part of the received speech coding parameters in the error detection section. By outputting the output signal as an output signal in that section, the received voice information can be reproduced with low quality even during the continuous error detection section, and the sense of discontinuity, which is a cause of auditory deterioration, can be eliminated. can get.

According to the second aspect of the present invention, in a continuous error detection section, a synthesis filter is formed by using all or a part of the received voice spectrum information of the error section, and the comfort filter is formed by using the synthesis filter. Is generated and output as an output signal of the section, the received phoneme information can be reproduced with low quality even during the continuous error detection section, and the sense of discontinuity, which is an auditory deterioration factor, can be eliminated. The effect is obtained.

In the invention according to the third aspect of the present invention, the spectrum quantization has a multi-stage configuration, and the scalar quantization which reduces the quantization efficiency but reduces the bit error effect in the first stage is performed in two stages. The configuration is such that the vector quantization with high quantization efficiency is performed on the eyes, and the decoded spectrum of only the first-stage scalar quantization part with high error resilience is comfortably obtained in the spectrum decoding unit of the comfortable noise generator in the continuous error section on the decoding side. By using it as a synthesized spectrum of noise,
The effect is obtained that spectral decoding with less influence of errors can be performed while having low quality.

Further, in the invention according to claim 4 of the present invention, when error detection is continued for a long period of time, the combined spectrum operates so as to approach the white noise spectrum, so that the error interval can be improved. The effect that natural error section signal reproduction can be performed is obtained.

Further, in the invention according to claim 5 of the present invention, when error detection is continued for a long period of time, the synthesized spectrum operates so as to approach the spectrum of the silent section, so that the past in the error section is obtained. Thus, an effect that a more natural error section signal can be reproduced having a spectrum characteristic similar to that of the reproduction signal in the silent section of the above can be obtained.

In the invention according to claim 6 of the present invention, the sound source similar to the reproduced signal of the past silent section is used in the error section by using the sound source signal of the past silent section as the sound source signal of the error detecting section. The effect is obtained that more natural error section signal reproduction having characteristics can be performed.

Further, in the invention according to claim 7 of the present invention, when the error detection is continuous, the output from the high-error-tolerant speech decoder is used as the decoded speech signal, so that the decoded speech is not interrupted. Can be reproduced.

Further, in the invention according to claim 8 of the present invention, the high-efficiency speech encoding by the high-efficiency speech encoder is performed by:
By performing the decoded speech signal of the high error resilience coding performed by the high error resilience encoder performed before that on the signal subtracted from the input speech, the decoded speech is not interrupted when the error detection is continuous. The effect is that the reproduction of the video and the reduction of the bit rate can both be achieved.

According to a ninth aspect of the present invention, there is provided a magnetic disk, a magneto-optical disk, a ROM cartridge, or the like, on which a program that implements the voice encoding device according to any one of the first to eighth aspects by software is recorded. , Or an apparatus that operates as an audio encoder using the recording medium, and has an effect that the audio encoding apparatus of the present invention can be realized by software.

[Brief description of the drawings]

FIG. 1 is a block diagram of an encoding side (a) and a decoding side (b) of a speech encoding apparatus according to Embodiment 1 of the present invention.

FIG. 2 is a block diagram of a comfort noise generator on the decoding side of the speech encoding device according to the second embodiment of the present invention.

FIG. 3 is a block diagram of a spectrum quantization unit (a) of a speech encoder on a coding side and a spectrum decoding unit (b) of a comfort noise generator on a decoding side in a speech encoding device according to a third embodiment of the present invention.

FIG. 4 is a block diagram of a spectrum decoding unit in a comfort noise generator on a decoding side of a speech encoding device according to a fourth embodiment of the present invention.

FIG. 5 is a block diagram of a spectrum decoding unit in a comfort noise generator on the decoding side of a speech coding apparatus according to a fifth embodiment of the present invention.

FIG. 6 is a block diagram of a noise source generator in a comfort noise generator on the decoding side in a speech coding apparatus according to a sixth embodiment of the present invention.

FIG. 7 is a block diagram of an encoding side (a) and a decoding side (b) of a speech encoding apparatus according to Embodiment 7 of the present invention.

FIG. 8 is a block diagram of a speech encoding apparatus according to Embodiment 8 of the present invention.

FIG. 9 is a block diagram of an encoding side (a) and a decoding side (b) of a conventional speech encoding apparatus.

[Description of Code] 101 speech encoder 102 error correction encoder 103 error correction / error detector 104 speech decoder 105 comfort noise generator 106 adder 201 speech parameter separation unit 202 spectrum decoding unit 203 noise source generation unit 204 Synthesis filter 205 Level control section 301 Scalar quantization section 302 Split vector quantization section 303 Scalar quantization decoding section 304 Split vector quantization decoding section 401 The section spectrum decoding section 402 White noise spectrum parameter buffer 403 Synthesis spectrum calculation section 501 The section Spectrum decoding section 502 Silent section spectrum parameter buffer 503 Synthetic spectrum calculating section 601 Silent section excitation signal buffer 602 Noise excitation section 701 High-efficiency speech encoder 702 Error correction encoder 703 High Error resilience encoder 704 Multiplexer 705 Demultiplexer 706 Error correction / error detector 707 High efficiency speech decoder 708 High error resilience speech decoder 709 Adder 801 Subtractor 901 Speech encoder 902 Error correction encoder 903 Error correction / error detector 904 Voice parameter interpolator 905 Voice decoder

Claims (9)

[Claims] 1. A speech coder for performing speech coding on an input speech and outputting a speech coding parameter on an encoding side, and an error correction coding and an error detection code for an obtained coding parameter. An error correction encoder that performs calculation and outputs transmission data; and on the decoding side, an error correction / error detector that performs error correction and error detection on the received data, and a speech coding parameter after error correction. A speech decoder that performs speech decoding, and a comfort noise generator that generates comfort noise using a part of speech coding parameters for an error detection section, in a section where error detection is continuous,
A speech coding apparatus that generates comfort noise using a part of the received speech coding parameters in the error detection section and outputs the comfort noise as an output signal in the section. 2. In speech encoding including at least spectrum information such as LPC parameters representing vocal tract information of input speech and sound source information corresponding to vocal cord information, the comfort noise generator on the decoder side includes: A speech spectrum decoding unit that decodes speech spectrum parameters using spectrum information among received speech coding parameters, a noise source generation unit that generates a noise source, and a synthesis filter that synthesizes and outputs a comfort noise signal. 2. The speech coding apparatus according to claim 1, wherein in a continuous error detection section, a synthesis filter is configured using all or a part of the received speech spectrum information of the error section, and comfort noise is generated using the synthesis filter. 3. A quantizer for a speech spectrum parameter in a speech encoder on the encoding side includes a scalar quantization unit for each order of the spectrum parameter in a first stage, and a plurality of spectrum parameters in a second stage and thereafter. Has a multi-stage configuration including a divided vector quantization unit that divides the vector into vector orders and performs vector quantization for each vector, and for an error detection section on the decoding side, in a spectrum decoding unit in a comfortable noise generator,
The speech coding apparatus according to claim 2, wherein the spectrum parameter is decoded using only the first-stage scalar quantization code among the received spectrum parameter codes. 4. A spectrum decoding unit in the comfort noise generator on the decoding side stores a section spectrum decoding unit for decoding a spectrum parameter from a received spectrum parameter code of the section and a spectrum parameter representing a white noise spectrum. A white noise spectrum buffer, and a synthesized spectrum calculation unit for generating a spectrum parameter for synthesizing a comfortable noise signal from the decoded spectrum parameter and the white noise spectrum parameter. 4. The device according to claim 2, which operates to asymptotically approach
A speech encoding device according to claim 1. 5. A spectrum decoding section in a comfort noise generator on a decoding side, wherein the spectrum decoding section decodes a spectrum parameter from a received spectrum parameter code of the section and a past silent section (including a section including only noise). ), A silence section spectrum parameter buffer storing spectrum parameters representing the spectrum of the spectrum, and a synthesized spectrum calculation section for generating a spectrum parameter for synthesizing a comfort noise signal from the decoded spectrum parameter and the silence section spectrum parameter. 4. The operation according to claim 2, wherein, when the detection is continuous, the synthesized spectrum is asymptotic to the spectrum of the past silent section.
A speech encoding device according to claim 1. 6. A noise source generating section in a comfort noise generator on a decoding side, comprising: a silent section source signal buffer for storing a source signal in a silent section; and a noise source generating section for generating a noise source using the buffer. The speech encoding apparatus according to any one of claims 2 to 5, further comprising using a source signal in a past silent section as a source signal in the error detection section. 7. On the encoding side, CELP is applied to input speech.
High-efficiency speech coder that performs high-efficiency speech coding at low bit rates such as coding, and error correction coding that performs error correction coding and error detection code calculation for the coding parameters obtained by the coding. And AD for input voice
A high error resilience encoder capable of realizing high error resilience despite low quality such as M coding, and performing error correction and error correction on the reception data corresponding to the output of the high efficiency encoder out of the reception data on the decoding side. An error correction / error detector for performing detection, a high-efficiency speech decoder for performing speech decoding corresponding to a high-efficiency speech coder from speech coding parameters after error correction, and a high error-tolerant speech code among received data. A high-error-tolerant speech decoder that performs speech decoding on received data corresponding to the output of the decoder, and switches or adds outputs from the high-efficiency speech decoder and the high-error-tolerant speech decoder according to the result of error detection. A speech encoding device comprising: an adder that outputs an output signal in the section; and when the error detection is continuous, an output from the high error-tolerant speech decoder is used as a decoded speech signal. 8. A coding apparatus further comprising a subtracter for subtracting a decoded speech signal of a high error resilience encoding by a high error resilience encoder from an input speech, and a high efficiency speech encoding by a high efficiency speech encoder. 8. The speech coding apparatus according to claim 7, wherein the decoding performs a decoded speech signal component of the high error resilience coding performed by the high error resilience coder performed prior to the signal subtracted from the input speech. 9. A recording medium such as a magnetic disk, a magneto-optical disk, a ROM cartridge, or the like, on which a program that implements the voice encoding device according to claim 1 by software is recorded, or using the recording medium. A device that operates as a speech coder.