JPH0728496A - Voice coder - Google Patents

Abstract

PURPOSE:To continue a normal operation of a voice coder even though it experiences an abnormal operating condition. CONSTITUTION:The voice coder is made of a DSP(digital signal processor) LSI and is provided with a residual signal generation section 1 in which voice signals are inputted as real number columns, residual signals that are the differences between the voice input signals and reproduced voice signals are obtained and outputted and a function is provided to generate the reproduced voice signals and a quantization section 2 which quantizes the outputted residuals of the section 1 and the parameters necessary for a decoding side. Furthermore, the voice coder is provided with a parameter storage section 4 which stores the initial setting values of the parameters used in the coder during a power supply is turned on and intermediate computation results that do not affect the operations of the coder for every auxiliary parameter made to realize the coder and a comparison discrimination conversion section 5 which compares the reproduced voice signals generated using the function of the section 1 and the voice input signals and converts the parameters and the auxiliary parameters to the values stored in the section 4 when the compared value exceeds a prescribed value.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coder which can operate normally thereafter even when it falls into an abnormal state and loses its original function.

[0002]

2. Description of the Related Art A method of digitizing and communicating a voice signal is being achieved by developing various voice encoders in order to increase the number of channels accommodated in the line and miniaturize the device. The lower the rate (64 kbps → 16 kbps or less), the more complicated the encoding process becomes, and the normal digital signal processor LSI
(DSP LSI).

As described above, a highly efficient speech coder uses the four arithmetic operations for sampling the features of speech as a sequence of real numbers. In a DSP LSI, a real value of a finite word length is represented by its own number of bits, and cannot be represented accurately when the absolute value of the operation result is very large or very close to zero.

Therefore, even if the encoding algorithm itself has no abnormality, depending on how the DSP program is made even if an abnormal state occurs due to these calculation accuracy, overload, and underload, or even if external noise does not occur. May fall into an abnormal state.

Since the internal state of the voice encoder is determined by some parameters (real values) determined by the encoding algorithm, the internal state may become abnormal due to external noise such as power supply fluctuation.

When the voice coder falls into an abnormal state, the subsequent operation is stopped. In order to continue normal operation after the speech coder has fallen into an abnormal state, it is conceivable to identify all abnormal states of the speech coder and restore the parameters in that case to the normal state.

[0007]

However, since the encoding algorithm is complicated, it is very difficult to identify all abnormal states, and normal operation continues after the voice encoder falls into an abnormal state. There is a problem that it is very difficult to do.

An object of the present invention is to provide a speech coder capable of continuing normal operation even after an abnormal state has occurred.

[0009]

FIG. 1 is a block diagram showing the principle of the present invention. As shown in FIG. 1, a voice signal is sampled by inputting voice features into a sequence of real numbers, a residual signal of a difference between the voice input signal and the reproduced voice signal is obtained and output, and a reproduced voice signal is generated. A residual signal generating unit 1 having a function capable of performing, a quantizing unit 2 for quantizing the residual signal output from the residual signal generating unit 1 and a parameter necessary for obtaining a reproduced voice signal on the decoder side. In the voice coder realized by the DSP LSI, the initial setting values of the parameters used in the voice coder when the power is turned on, and the parameters to be supplementarily created in realizing the voice coder in the DSP LSI. , A parameter storage unit 4 for storing a calculation result in the middle that does not hinder the operation of the subsequent speech encoder, and a reproduced voice signal generated by using the function of the residual signal generation unit 1 and the voice input signal. Is different from the specified value When is configured to include a comparison determination converter 5 for converting a value which stores the parameters and the auxiliary parameters in the parameter storage unit 4.

[0010]

According to the present invention, attention is paid to the fact that the reproduced voice signal greatly differs from the voice input signal when an abnormal state occurs, and the comparison / determination conversion unit 5 generates the residual signal using the function of the residual signal generation unit 1. When the reproduced voice signal and the voice input signal are compared, and when they differ by a predetermined value or more, the parameters and the auxiliary parameters stored in the parameter storage unit 4 are initialized when the power is turned on for the parameters used in the voice encoder. Value and DSPLSI
In order to realize a speech coder, it is converted to the value of the calculation result in the middle that does not hinder the operation of the speech coder after that for each parameter that is supplementarily created. It will work.

[0011]

2 is a block diagram of a speech encoder and decoder according to an embodiment of the present invention. In FIG. 2, the audio signal is sampled into a sequence of real values by inputting the audio features and is input to the adder 11 and the predictor 10.

The prediction unit 10 predicts the present voice input signal from the past voice input signal, obtains the predicted signal, and the addition unit 1
1 and subtracts the prediction signal from the voice input signal to obtain the residual signal, which is input to the parameter quantizer 12.

Further, some parameters generated in the prediction by the prediction unit 10 are also input to the parameter quantization unit 12 and quantized. At this time, the residual signal generation unit 14 based on the quantization parameter. The prediction signal generated by the prediction unit 15 based on the generated reproduction residual signal and the quantization parameter is also used (or may not be used), and the parameter quantized by the parameter quantization unit 12 depends on the quantization parameter. In addition to being used by the residual signal generation unit 14 and the quantization parameter prediction unit 15, it is sent to the coding code multiplexing unit 13 so that it can be reproduced on the decoder side, multiplexed and sent to the decoder via the transmission path.

The addition unit 16 adds the reproduction residual signal generated by the quantization parameter residual signal generation unit 14 and the prediction signal generated by the quantization parameter prediction unit 15 to generate a reproduction audio signal. obtain.

In the case of the present invention, the comparing section 5-1 and the judging section 5
2, a parameter conversion unit 5-3, and a parameter storage unit 4 are added, and the comparison unit 5-1 takes in and compares the audio input signal and the reproduced audio signal (reproduction signal) and compares the result with the judgment unit 5-2. send.

When the comparison result differs by a predetermined value or more, the judging section 5-2 judges that the state is abnormal, and the parameter converting section 5-3 stores the parameter and the auxiliary parameter in the parameter storage section 4. , The parameter initial setting value at power-on, and the value of the calculation result in the middle that does not hinder the subsequent operation of the speech coder for each parameter that is supplementarily created in realizing the speech coder with the DSP LSI. Convert.

Therefore, the operation continues normally thereafter. still,
The coded code sent to the decoder through the transmission line is input to the coded code separation unit 17, the same quantization parameter as that on the encoder side is created, and the residual signal generation unit 19 and the quantum based on the quantization parameter are used. It is input to the prediction unit 18 based on the quantization parameter, the residual signal generation unit 19 based on the quantization parameter generates a reproduction residual signal, and is input to the prediction unit 18 based on the quantization parameter and the addition unit 20.

The quantizing parameter prediction unit 18 generates a prediction signal and inputs it to the addition unit 20 to add the reproduction residual signal to obtain a reproduction voice signal. FIG. 3 is a block diagram of a speech coder and a decoder according to another embodiment of the present invention.

The speech coder of FIG. 3 is a kind of code excitation linear predictive coder and corresponds to the case where the quantization parameter is only the residual signal. In FIG. 3, the voice signal is sampled as a sequence of real numbers by sampling the voice feature, is input to the buffer 21, is stored for one frame = n samples, and the output is input to the adder 27.

Minimum error judgment and codebook selection unit 28
Selects a code that minimizes the residual signal from the excitation codebook 22 that includes m codes that simulate n samples of the residual signal obtained by subtracting the reproduced audio signal from the audio input signal. The index is sent to the decoder via the transmission line.

The reproduced voice signal is obtained from the code of the excitation codebook 22 through the gain section 23 and the synthesis filter 25. The adjustment of the gain of the gain section 23 is performed by the reproduction residual corresponding to the code actually selected in the past. The difference gain signal is used in the backward gain adaptor 31, and the reproduction voice signal generation of the synthesis filter 25 is performed by the backward synthesis filter adaptor 33 using the reproduction voice signal corresponding to the code actually selected in the past. Be seen.

In the case of the present invention, a comparison / determination unit 5-4, a parameter conversion unit 5-3, and a parameter storage unit 4 are added, and the comparison / determination unit 5-4 compares the audio input signal with the reproduced audio signal. When the comparison result is different by a predetermined value or more, it is determined as an abnormal state, and the parameter conversion unit 5-3 stores the parameter and the auxiliary parameter in the parameter storage unit 4 and the initial setting value of the parameter when the power is turned on. Then, each value is converted into a value of a calculation result on the way that does not hinder the subsequent operation of the speech coder for each parameter that is supplementarily created when the speech coder is realized by the DSP LSI.

Therefore, the operation continues normally thereafter. still,
In the decoder, the excitation codebook 29 selects a code from the index of the code sent via the transmission line,
From the selected code, the gain unit 30 and the backward gain adaptor 3 are selected.
1, the synthesis filter 32, and the backward synthesis filter adaptor 33 are used to obtain and output the reproduced speech signal in the same manner as in the speech coder.

FIG. 4 is a block diagram of a speech coder according to another embodiment of the present invention, and FIG. 5 is a block diagram of a decoder of the speech coder of FIG. The speech coder of FIG. 4 is also a type of code excitation linear predictive coder.

In FIG. 4, the high-pass filter 40 removes the direct current component from the audio signal which is obtained by sampling the characteristics of the audio into a sequence of real-valued data, and the buffer 41 outputs one frame =
The frame energy quantizing unit 42 calculates energy (power) for the stored signals for n samples, and the multiplexing unit 55 calculates the quantized frame energy.
To enter.

The prediction coefficient obtained by performing the linear prediction analysis in the linear prediction analysis section (LPC analysis section) 43 is quantized in the prediction coefficient quantization section 44, and the quantized prediction coefficient is sent to the multiplexing section 55. Together with the one-frame signal stored in the buffer 41, the residual signal P is sent to the prediction residual signal calculation unit 45.
Is calculated, and the gain vector quantization unit 46 and the lag search unit 4 are calculated.
7. Send to the codebook search unit 48.

The reproduction residual signal close to the residual signal P is calculated, and the reproduction residual signals for the past m samples are stored in the m sample buffer 49. The lag searching unit 47 selects consecutive n samples having the minimum error from the residual signal P from among the reproduced residual signal m samples, and selects them.
And The position of n samples in the m samples corresponds to the bitch cycle of the voice, and is used as a lag by the multiplexing unit 55.
Sent to.

In order to approximate the residual signal P, k basis vectors (fixed continuous n samples) q'i which are the source of the noise component are used in addition to the past samples. Q'in the orthogonal part 50
i is orthogonal to b, and the codebook search unit 48 calculates
Among the 2 ^k combinations Σ (±) _i × qi, the one with the smallest error from the residual signal P is selected. The selected one is designated as f and the code is sent to the multiplexing unit 55 as a code word.

In the gain vector quantizer 46, the one having the smallest error with the residual signal P among the linear combinations (β × b + γ × f) j of b and f is selected. At this time, the frame energy is quantized in j ways and the result is sent to the multiplexing unit 55 as a Geny vector code.

From the coefficients β and γ quantized by the gain vector quantization unit 46, the reproduction residual signal β × b is calculated by the calculation unit 51.
+ Γ × [Σ (±) _i × q'i] is obtained. Then, the multiplexing unit 55 multiplexes the quantized frame energy, the prediction coefficient, the code word, the lag, and the Geny vector code, and sends them to the decoder via the transmission path.

In the case of the present invention, a synthesizing filter 52, a comparison / decision section 5-4, a parameter converting section 5-3, and a parameter storing section 4 are added and reproduced by the synthesizing filter 52 using the reproduction residual signal and the prediction coefficient. A voice signal is obtained, a voice input signal and a reproduced voice signal are compared by a comparison / determination unit 5-4, and when the comparison result is different by a predetermined value or more, it is determined as an abnormal state, and a parameter conversion unit 5-3 determines a parameter and Auxiliary parameters are stored in the parameter storage unit 4 and initial setting values at the time of power-on of the parameters, and parameters of the following audio encoder for each auxiliary parameter to be implemented in realizing the audio encoder with the DSP LSI. It is converted to the value of the calculation result in the middle that does not hinder the operation.

Therefore, the operation continues normally thereafter. still,
In the decoder shown in FIG. 5, the signal sent from the speech coder via the transmission path is separated into a frame energy, a prediction coefficient, a code word, a lag, and a Genie vector code by the separation unit 60, and the frame energy is calculated. The Geny vector code is input to the Geny vector decoding unit 61, the coefficients γ and β are calculated and input to the calculation unit 66, the prediction coefficient is input to the synthesis filter 65, and f described in the case of the encoder is set to the Geny vector. The code word is input to the decoding unit 61, the code word is input to the excitation code word decoding unit 62, the output is input to the arithmetic unit 66, the lag is input to the lag decoding unit 63, and the code from the buffer 64 for m samples is input. , Which is described in the case of the encoder, is input to the Geny vector decoding unit 61 and the arithmetic unit 66, and the arithmetic unit 66 performs the same arithmetic operation as the encoder to obtain a reproduction residual signal and a synthesis filter. Input to 65 and synthesize And outputs to obtain a reproduced audio signal by the filter 65.

[0033]

As described in detail above, according to the present invention, when the voice coder falls into an abnormal state and cannot perform its original function, the parameters and the auxiliary parameters are initially set when the power is turned on. It changes to the value and the value of the calculation result in the middle of not interfering with the operation of the subsequent speech encoder for each parameter that is supplementarily created in realizing the speech encoder with DSP LSI, and after that it operates normally. Has the effect of continuing.

[Brief description of drawings]

FIG. 1 is a block diagram of the principle of the present invention,

FIG. 2 is a block diagram of a speech encoder and a decoder according to an embodiment of the present invention,

FIG. 3 is a block diagram of a speech encoder and decoder according to another embodiment of the present invention,

FIG. 4 is a block diagram of a speech encoder according to another embodiment of the present invention,

FIG. 5 is a block diagram of a decoder of the speech encoder of FIG.

[Explanation of symbols]

 1 is a residual signal generation unit, 2 is a quantization unit, 3 is a reproduction signal generation unit, 4 is a parameter storage unit, 5 is a comparison / determination conversion unit, 5-1 is a comparison unit, 5-2 is a determination unit, 5− 3 is a parameter conversion unit, 5-4 is a comparison determination unit, 10 is a prediction unit, 11, 16, 20, and 27 are addition units, 12 is a parameter quantization unit, 13 is an encoded code multiplexing unit, and 14 and 19 are quantum units. Residual signal generation section by quantization parameter, 15 and 18 prediction section by quantization parameter, 17 coded code separation section, 21, 41, 49 and 64 buffers, 22 and 29 excitation codebook, 23 and 30 Gain unit, 24 and 31 are backward gain adaptors, 25, 32, 52 and 65 are synthesis filters, 26 and 33 are backward synthesis filter adaptors, 28 is a minimum error judgment and codebook selection unit, 40 is a high pass filter, 42 Is Energy quantization unit, 43 linear prediction analysis unit, 44 prediction coefficient quantization unit, 45 prediction residual signal calculation unit, 46 gain vector quantization unit, 47 lag search unit, 48 codebook search unit, 50 is an orthogonal unit, 51 and 66 are arithmetic units, 55 is a multiplexing unit, 60 is a separating unit, 61 is a gain vector decoding unit, 62 is an excitation codeword decoding unit, and 63 is a lag decoding unit. .

Claims (1)

[Claims] 1. A voice signal can be input by sampling a voice feature and inputting a voice signal in a sequence of real-valued values, obtaining and outputting a residual signal of a difference between a voice input signal and a playback voice signal, and generating a playback voice signal. Residual signal generating section (1) having a function, quantization for quantizing a residual signal output from the residual signal generating section (1) and a parameter necessary for obtaining a reproduced audio signal on the decoder side In a voice coder realized by a digital signal processor LSI having a section (2), initial setting values of power-on parameters of parameters used in the voice coder and the digital signal processor LSI For each parameter that is supplementarily created in realizing a speech coder,
A parameter storage unit (4) for storing a calculation result in the middle that does not hinder the operation of the subsequent speech encoder, a reproduced speech signal generated by using the function of the residual signal generation unit (1), and the speech input signal. And the parameter storage unit (4) stores the parameter and the auxiliary parameter when they differ by more than a predetermined value.
A speech coder, comprising: a comparison / determination conversion unit (5) for converting into a value stored in.