JP2968109B2 - Code-excited linear prediction encoder and decoder - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an encoder and a decoder according to a code-excited linear prediction encoding method (CELP).

[0002]

2. Description of the Related Art Conventionally, as a high-efficiency coding method for voice signals (including audio signals) in the field of digital mobile communication,
A code-excited linear predictive coding scheme (VSELP), which is a modification of the code-excited linear predictive coding scheme, has been adopted.

The basic configuration of a coding method for a speech signal is to obtain a vocal tract parameter representing vocal tract characteristics of speech and a sound source parameter representing sound source information. In recent code-excited linear predictive coding, the excitation signal as the sound source information is divided into an adaptive code that contributes to a voiced sound with a statistically strong periodicity and a statistical code that contributes to a random unvoiced sound with a statistically weak periodicity. And store them in a codebook, and find the optimal adaptive code and statistical code in each codebook so that the sum of weighting error power between the input audio signal and the synthesized audio signal is minimized. Performs the encoding process. And at least a sound source parameter, whether it is a forward type encoding method for obtaining vocal tract parameters from an input audio signal or a backward type encoding method for obtaining vocal tract parameters from a synthesized audio signal.
Therefore, the information of the optimal adaptive code and statistical code is transmitted.

It is known that such a code excitation linear predictive coding system can provide high-quality reproduced speech at a coding speed of 6 kbit / s to 8 kbit / s.

[0005]

However, some communication systems require a lower coding rate. For example, there is one that calculates an encoding speed of 4 kbit / s or less. In the case of such a low speed, whether the forward type transmits both the vocal tract parameters and the sound source parameters,
Regardless of the backward type that transmits sound source parameters,
Naturally, the number of coding bits allocated to the excitation parameters is reduced, and the number of adaptive codes and statistical codes stored in the adaptive codebook and statistical codebook is also reduced. As a result, at such a low encoding speed, the quality of the reproduced sound is reduced.

Further, since the adaptive codebook is updated adaptively with a combination code of an optimal adaptive code and a statistical code, it can be said that the adaptive code is formed based on the statistical code. . Therefore, there is a disadvantage that a voiced sound having strong periodicity rises slowly, and a code having a strong pulse characteristic cannot be formed even in a stationary portion of the voiced sound, and the reproduced voice lacks clarity. To overcome this disadvantage, a pulse consisting of an isolated impulse
It is also conceivable to use security codes.

However, a statistical code or a pulse code
Code is a fixed code, let's improve the quality of playback audio
Is a fixed code such as a statistical code or pulse code
The types must be very many. But this
In such a case, it is impossible to cope with a low encoding speed.

On the other hand, if the fixed code is reduced, the input sound
It becomes difficult to obtain a reproduced sound that accurately reflects the voice signal. example
For example, the characteristics of vocal cords (sources of sound source information) differ depending on the individual.
However, such differences in vocal cords are corrected in the excitation signal.
And the quality of the playback audio is reduced.
You.

The present invention has been made in view of the above points, and has a code-excited linear prediction method capable of improving the quality of reproduced speech mainly from a sound source parameter plane even at a low coding rate. It is intended to provide an encoder and a decoder.

[0010]

Means for Solving the Problems In order to solve the problems, the first invention is as follows: (1) Linear prediction of an input speech signal
Vocal tract parameter creation means for determining vocal tract parameters by
(2) The vocal tract parameters are quantized and quantized
Amount of vocal tract parameters that output at least vocal tract parameters
(3) optimal code of past input audio signal
Adaptive codebook for storing an adaptive code representing the
And (4) a predetermined statistical code and / or pal
Fixed code that stores a fixed code consisting of security codes
Book, and (5) based on the adaptive code and the fixed code.
Excitation signal creating means for creating an excitation signal based on the
(6) The quantized vocal tract parameters and the excitation signal
And a synthesized speech signal is created based on the input speech signal.
And evaluating the error between the synthesized speech signal and
Supports the best code for the input audio signal at the current time
Determine the index of the two codebooks, and
Parameters and the above indices are coded audio signals.
In a code-excited linear prediction encoder that outputs
(7) The excitation signal creating means includes the vocal tract parameter
Create a specific impulse response based on
Convolution processing of the above fixed code using the
After that, the excitation signal is created by combining with the adaptive code.
(8) the impulse response is fixed
The input audio signal determined at the time the code was output
Of the transfer function that converts to the frequency response of
It is characterized by being responsive.

Further , the second invention provides (1) code excitation
The encoded audio signal given from the linear prediction encoder is
Vocal tract parameters, adaptive code index and fixed code
Demultiplexing means for demultiplexing into code indexes, (2)
Past input corresponding to the index of the above adaptive code
An adaptive code that outputs a speech signal code as an adaptive code.
And (3) the index of the above fixed code.
Fixed to output a predetermined fixed code corresponding to
A code book, (4) the adaptive code and the fixed code
Excitation signal generating means for generating an excitation signal based on the
And (5) based on the vocal tract parameters and the excitation signal
Means for creating a synthesized speech signal by
(6) The code-excited linear prediction decoder
The excitation signal generating means is configured to generate a characteristic based on the vocal tract parameters.
Create a constant impulse response and use the impulse response
After performing the convolution process on the fixed code,
The excitation signal is created by adding to the adaptive code.
(7) The fixed code is output from the impulse response.
The frequency characteristics of the input audio signal determined at the time
Corresponding to the impulse response of the transfer function to be converted
It characterized in that it is.

[0012]

According to a first aspect of the present invention, there is provided a code-excited linear predictive encoder and a code-exciting linear predictive encoder.
And a second embodiment of the code-excited linear prediction decoder according to the present invention.
Generates a specific impulse response based on vocal tract parameters
Create a fixed code using the impulse response.
After convolving the code with the adaptive code,
And the impulse response is
The above input sound determined at the time when the fixed code is output
Impulse response of transfer function converted to frequency characteristics of signal
Therefore, it is an element of the excitation signal.
Frequency characteristics of the input audio signal
And a number of fixed codes are available.
Obtain high quality playback audio even if it is not
Can be.

That is, the output from the fixed codebook is
An operation is performed to bring the frequency characteristics of the fixed code closer to the frequency characteristics of the input audio signal. The reason for this is that the frequency characteristic of the excitation signal has been theoretically modeled as white in the past, but it is not white in nature but has a characteristic close to the frequency characteristic of the input audio signal. Has been experimentally confirmed, such as statistical codes and pulse codes
This is because, if the frequency characteristics of the fixed code are closer to the frequency characteristics of the input voice signal, a higher quality synthesized voice signal can be obtained, and the effective frequency component of the excitation signal is the quantization error signal. This is because the masking effect becomes much larger and a masking effect of the quantization error signal is obtained.

[0014]

【Example】

(A) One Embodiment of Code Excited Linear Prediction Encoder An embodiment of a code excited linear prediction encoder according to the present invention will be described in detail below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of this embodiment.

Referring to FIG. 1, the code-excited linear predictive encoder of this embodiment comprises an input speech processing section 1, an optimum synthesized speech search section 2, and a multiplexing section 3.

The input speech processing unit 1 includes an LSP (line spectrum vs. parameter) analyzer 11, an LSP parameter encoder 12, an LSP parameter decoder 13, an LPC (linear prediction coefficient) coefficient converter 14, and a weighting filter 15. , A synthesis filter zero-input response generator 16, a weighting filter zero-input response generator 17, and two subtractors 18 and 19, and when an input audio signal is given, is transmitted to a decoder. The vocal tract parameters are obtained, and a target audio signal of a synthesized audio signal formed by local reproduction is formed.

In this embodiment, a digitized discrete input speech signal sequence is accumulated for a time corresponding to an analysis frame length for obtaining vocal tract parameters, and the analysis frame length is several times. It is divided into sub-frames and processed by the input audio processing unit 1.

The input voice signal is supplied to an LSP analysis unit 11, which performs an LSP analysis and converts it into LSP parameters as vocal tract parameters.
This LSP parameter is the LSP parameter encoding unit 12
(For example, vector quantization), and the resulting signal is supplied to the multiplexing unit 3 and transmitted to the code excitation linear prediction decoder. Also, the encoded LSP parameter is LS
After being decoded (vector inverse quantization) by the P parameter decoding unit 13, the LPC coefficient conversion unit 14
It is converted to a C coefficient. The LPC coefficients thus converted are used as tap coefficients of the weighting filter 15, the synthesis filter zero-input response generator 16, the weighting filter zero-input response generator 17, and a weighting synthesis filter 29 described later. It is also provided to a frequency characteristic operation unit 28 described later. Note that, instead of directly converting the LSP parameters output from the LSP analysis unit 11 into LPC coefficients, codes,
The conversion of the decoded LSP parameters into LPC coefficients is performed in the same manner as the LPC coefficients used by the decoder.
This is because the sound source parameters can be appropriately determined by using the C coefficient in local reproduction.

Here, the LSP parameter is used as the vocal tract parameter to be used for transmission because the interpolation characteristic with respect to the frequency characteristic of the vocal tract is improved, and even if the LSP parameter is encoded with a small number of encoding bits, the LPC parameter is used. This is because distortion given to the vocal tract spectrum is smaller than parameters and the like, and efficient coding can be performed by combination with the vector quantization method.

Next, the operation of forming a target audio signal for a synthesized audio signal locally reproduced from an input audio signal will be described.

The input audio signal described above is applied to the weighting filter 1
5, weighted in consideration of human auditory characteristics, and then supplied to the subtractor 18 as a subtracted input.
The subtracter 18 is supplied with a zero input response signal relating to the weighted synthesis filter 29 generated by the synthesis filter zero input response generation unit 16 using the LPC coefficient as a tap coefficient, as a subtraction input. Thus, an audio signal from which the influence of the state of the weighted synthesis filter 29 in the immediately preceding analysis frame has been removed is obtained, and this is supplied to the subtractor 19 as an input to be subtracted. The subtractor 19 also includes a weighting filter 1 generated by the weighting filter zero-input response generator 17 using LPC coefficients as tap coefficients.
The quiescent response signal for 5 is provided as the subtraction input. Thus, an audio signal from which the influence of the state of the weighting filter 15 in the immediately preceding analysis frame has been removed is obtained.
This is supplied to a subtractor 30 described later as a target audio signal.

The optimum synthesized speech search unit 2 searches for a sound source parameter whose synthesized speech signal obtained by local reproduction is most similar to the target speech signal, and includes an adaptive codebook 20, a statistical codebook 21, a pulse codebook 22, Gain codebook 23, gain controllers 24 and 27, adder 2
5, fixed code selection switch 26, frequency characteristic operation unit 2
8, a weighted synthesis filter 29, a subtractor 30, an error power sum calculator 31 and a minimum error power sum code selector 32.

The adaptive codebook 20, the statistical codebook 21, and the pulse codebook 22 store an adaptive code, a statistical code, and a pulse code, which are waveform codes related to an excitation signal, respectively. Reference numeral 23 stores a gain code relating to the adaptive code and the fixed code (the statistical code and the pulse code are collectively referred to as such).

The adaptive code and the statistical code are respectively a waveform code contributing to a voiced sound having a statistically strong periodicity and a waveform code contributing to a random unvoiced sound having a statistically weak periodicity. In addition, adaptive code book 2
The adaptive code of 0 is adaptively updated as described later.
The pulse code is a waveform code composed of an isolated impulse. The pulse-like code takes into account the fact that it contributes to the rise of a voiced sound with a strong periodicity and to the stationary part of a voiced sound with a clear pulse. The gain code is, for example, vector-quantized, and one component of the vector relates to the gain of the adaptive code, and the other component relates to the gain of the fixed code.

Since the pulse-like sound source signal is a simple signal having a periodicity, a pulse signal generator may be generated. However, the pulse-like sound source signal is coded and read out from the code book 22 as in this embodiment. This is preferable for the following reasons. That is, it is easy to synchronize with the output from the adaptive codebook 20, and by using the same book configuration as the statistical codebook 21, the statistical code or the pulse code is selected and transmitted to the decoder as described later. This is because multiplexing processing and the like can be easily performed.

The optimum code of each code, in which the synthesized voice signal locally reproduced using such various codes is most similar to the target voice signal, is obtained, and its index is given to the multiplexing unit 3 to perform code excitation linear predictive decoding. To the transmitter. In this embodiment, since a low coding rate is considered, a statistic code or a pulse code is selected for a fixed code and the index thereof is transmitted. Therefore, the selection information of which one is selected as the fixed code is also transmitted to the code excitation linear prediction decoder side.

In this embodiment, the search for the optimum code (including the selection of the statistical code or the pulse code) is performed in the order of the adaptive code, the statistical code, the pulse code, and the gain code.

At the time of searching for an optimal adaptive code, the outputs from the statistical codebook 21 and the pulse codebook 22 are set to 0, and the gain controller 24 multiplies an appropriate value of a gain coefficient (for example, 1). It has been made like that. In such a state, all the adaptive codes stored in the adaptive code book 20 are output in time sequence or in parallel, and are output to the weighted synthesis filter 29 via the gain controller 24 and the adder 25 as excitation signals. give. The weighted synthesis filter 29 performs convolution processing on the excitation signal using the LPC coefficient given from the LPC coefficient conversion unit 14 as a tap coefficient, and generates a synthesized speech signal in which only the content of the adaptive code is reflected as a sound source parameter, Obtain for all adaptive codes.

The subtracter 30 obtains an error signal between the synthesized speech signal in which only the content of the adaptive code is reflected and the target speech signal for all the adaptive codes, and supplies it to the error power sum calculator 31. The error power sum calculator 31 calculates the sum of squares (error power sum) of the components of the error signal for all the adaptive codes and supplies the sum to the minimum error power sum code selector 32.
The minimum error power code selection unit 32 determines the adaptive code having the minimum error power sum as the optimal one.

Next, a search for an optimum statistical code is performed. In this search, the fixed code selection switch 26 is switched to the statistical codebook 21 side, and the output of the adaptive codebook 20 is set to 0 ( It is also conceivable to output an optimal adaptive code). In such a state, all the statistical codes stored in the statistical code book 21 are output in time sequence or in parallel, and input to the frequency characteristic operation unit 28 via the fixed code selection switch 26 and the gain controller 24. Let it.

The frequency characteristic operation unit 28 has a detailed configuration shown in FIG. 3 described later, and converts the frequency characteristics of the input statistic code into the frequency characteristics of the input audio signal corresponding to the time length of the statistic code. Perform the conversion operation to approach. All the statistical codes whose frequency characteristics have been converted in this way are supplied to the weighted synthesis filter 29 as an excitation signal via the adder 25 (equivalently in this case). Subsequent processing is performed in the same manner as the search for the optimal adaptive code, and the minimum error power sum code selection unit 32 determines the optimal statistical code.

The reason why the frequency characteristic operation section 28 is provided is as follows. Conventionally, the frequency characteristics of the excitation signal have been theoretically modeled as white, but it has been experimentally confirmed that the frequency characteristics of the excitation signal are not white and have characteristics close to the frequency characteristics of the input audio signal. I have. Therefore, the closer the frequency characteristics of the statistical code or pulse code are to those of the input audio signal, the higher the quality of the synthesized audio signal can be obtained, and the effective frequency components of the excitation signal are quantized. It becomes much larger than the error signal, and a masking effect of the quantization error signal is obtained. Therefore, a frequency characteristic operation unit 28 is provided. Here, the information representing the frequency characteristics of the input speech signal includes an LPC coefficient, and information on an optimal adaptive code (including a gain for the code), which means pitch prediction information. Therefore, the frequency characteristic operation unit 28 operates the frequency characteristics of the statistical code or the pulse code based on the information.

When the search for the optimum statistical code is completed in this way, the search for the optimum pulse code is next performed. At the time of this search, the fixed code selection switch 26 is switched to the pulse codebook 22 side, and the output of the adaptive codebook 20 is set to 0 (note that the optimum adaptive code may be output). In such a state, all the pulse codes stored in the pulse code book 22 are output in time sequence or in parallel. Subsequent processing is the same as that at the time of searching for the optimum statistical code, and therefore, description thereof is omitted.

When the optimum pulse code is determined in this way, the minimum error power sum code selector 32
Compares the error power sum of the optimal statistic code and the error power sum of the optimal pulsed code, and determines the smaller one as the fixed code to be transmitted to the code excitation linear prediction decoder.

Thereafter, a search for an optimum gain code is performed. During the search for the gain code, the adaptive codebook 20 outputs the optimal adaptive code, and the fixed code selection switch 26 sets the selected statistical codebook 21.
Alternatively, the codebook is switched to the pulse codebook 22, and the selected fixed codebook 21 or 22 outputs the optimum fixed code. One gain codebook 23 includes an adaptive code gain and a fixed code gain. The adaptive code gain is provided to a gain controller 24, and the fixed code gain is provided to a gain controller 27. Thus, the optimal adaptive code subjected to the gain control and the optimal fixed code subjected to the frequency characteristic operation and the gain control are added by the adder 25, and given to the weighted synthesis filter 29 as an excitation signal. Such processing is performed by the gain codebook 23.
Is performed in time sequence or in parallel for all the gain codes in. The processing at the time of searching after the weighted synthesis filter 29 is the same as the processing at the time of searching for other codes.

When the optimum adaptive code, the optimum fixed code, and the optimum gain code are obtained, the minimum error power sum code selecting unit 32 gives these indices to the multiplexing unit 3 and selects either the statistical code or the pulse code. The fixed code selection switch information indicating the selection is also given to the multiplexing unit 3. The multiplexing unit 3 multiplexes the LSP parameters supplied from the LSP parameter encoding unit 12 and the information, and outputs the multiplexed information to the code excitation linear prediction decoder side. When vector quantization is applied as a gain code, the transmitted index is a vector number.

The minimum error power sum code selector 32
Stores the index and fixed code selection switch information given to the multiplexing unit 3 in the corresponding codebook (20 and 2).
3 and 21 or 22) and the fixed code selection switch 26. At this time, the switch 26 is switched, and the optimal code is output from each codebook. As a result, an excitation signal capable of forming a synthesized voice signal closest to the target voice signal in the current subframe processing is output from the adder 25, and supplied to the adaptive codebook 20.
Then, adaptive code book 20 performs an adaptive code update process.

The above-described encoding process is repeated for each subframe, and the encoded speech signal is sequentially transmitted to the code excitation linear prediction decoder.

FIG. 3 shows a detailed configuration of the frequency characteristic operation unit 28 described above. In FIG. 3, the frequency characteristic operation unit 28 includes two filters 28 connected in cascade.
1 and 282, and a pitch lag determining unit 283.

The fixed code output from the fixed code selection switch 26 is provided to the first filter 281.
The impulse response H 1 (z) of the first filter 281
Is selected as shown in the following equation, and performs a frequency conversion operation on the input fixed code.

H 1 (z) = (1−Σγ ⁱ aiz ⁻ⁱ ) / (1−Σν ⁱ aiz ⁻ⁱ ) (1) where ai (i is 1 to M) is supplied from the LPC coefficient conversion unit 14. The tap coefficient for the synthesis filter 29 to be used. Further, γ and ν are each from a predetermined 0.
It is a constant that is largely smaller than 1.

The fixed code whose frequency characteristics have been manipulated by the first filter 281 is
Is input to The pitch lag determining unit 283 obtains the pitch lag L from the index of the optimal adaptive code for the adaptive codebook 20, and supplies the pitch lag L to the second filter 282. The impulse response H2 (z) of the second filter 282 is
The frequency is selected as shown in the following equation, and a frequency conversion operation is performed on the input fixed code.

H2 (z) = 1 / (1−εz− ^L ) (2) where ε is a constant greater than 0 and less than or equal to 1 which is predetermined. The output of the second filter 282 is provided to the gain controller 27.

As described above, the frequency characteristic of the input fixed code is converted into the frequency characteristic of the input audio signal in accordance with the time length of the fixed code by the frequency characteristic operation unit 28 having such a detailed configuration. It is designed to get closer.

Therefore, according to the code-excited linear predictive encoder of the above-described embodiment, high-quality reproduced speech can be obtained even at a low encoding speed. Hereinafter, it will be specifically described that such effects can be obtained.

(1) When a low coding rate is expected, the number of coded bits allocated to the excitation parameters is small, so that the number of fixed codes to be prepared is also small, and the pulse noise included in the input speech signal can be clearly identified. However, in this embodiment, since the pulse code is used, the sound reproduction quality in such a case can be improved.

Further, since the pulse code and the statistical code are switched and used, it is possible to cope with a low encoding speed and to improve the reproduction quality of a signal in which a random signal and a pulse signal are mixed, such as a transient part of voice. Can be enhanced.

(2) When a low encoding speed is expected, the number of encoded bits for the excitation parameters is reduced, but the number of encoded bits for the vocal tract parameters is also reduced. In the case of this embodiment, even if encoding is performed with a small number of encoding bits, LP
Since the LSP parameter that causes less distortion to the vocal tract spectrum than C or the like is transmitted, the reproduction quality can be improved from this point.

(3) As described above, the frequency characteristic operation unit is provided in consideration of the fact that the actual excitation signal has a frequency characteristic close to the frequency characteristic of the input audio signal. Only while improving the playback quality,
With this conversion, a masking effect on the quantization error signal is generated, and the reproduction quality can be improved.

(B) One Embodiment of Code Excited Linear Predictive Decoder Next, an embodiment of a code excited linear predictive decoder according to the present invention will be described in detail with reference to the drawings. This embodiment corresponds to the embodiment of the code excitation linear prediction encoder shown in FIG. FIG. 2 is a block diagram showing the configuration of this embodiment.

Referring to FIG. 2, a code excitation linear prediction decoder according to this embodiment includes a demultiplexer 40, an LSP parameter decoder 41, an LPC coefficient converter 42, an adaptive codebook 43, a statistical codebook 44, Codebook 45, gain codebook 46, gain controllers 47, 49,
Fixed code selection switch 48, frequency characteristic operation unit 50,
It comprises an adder 51 and a weighted synthesis filter 52.

The coded speech signal given from the code excitation linear prediction coder side is input to the demultiplexing section 40.
The demultiplexing unit 40 separates the encoded audio signal into LSP parameters, an index of an optimal adaptive code, an index of an optimal fixed code, an index of an optimal gain code, and fixed code selection switch information. And LS
The P parameter is given to the LSP parameter decoding unit 41,
The index of the optimal adaptive code is stored in the adaptive code book 43.
, The index of the optimal gain code is provided to the gain codebook 46, and the fixed code selection switch information is provided to the fixed code selection switch 48. Further, the index of the optimal fixed code is given to the statistical codebook 44 or the pulse codebook 45 determined based on the fixed code selection switch information.

The LSP parameter decoding section 41 decodes (for example, vector dequantizes) the given coded LSP parameters, and the LPC coefficient conversion section 42 converts the LSP parameters into LPC coefficients. The LPC coefficients thus converted are provided to the frequency characteristic operation unit 50 and the weighted synthesis filter 52 as vocal tract parameter information.

The gain codebook 46 separates a gain code determined by a given index into an adaptive code and a fixed code, and sends them to an adaptive code gain controller 47 and a fixed code gain controller 49, respectively. give.

The adaptive codebook 43 outputs an adaptive code determined by a given index, and the adaptive code is gain-controlled via a gain controller 47 and is applied to an adder 51. The adaptive code book 43 gives the adaptive code to the frequency characteristic operation unit 50.

The statistical code book 44 or the pulse code book 45 stores the statistical code or the pulse code corresponding to the given index in the fixed code selection switch 4.
8 to the frequency characteristic operation unit 50, and the frequency characteristic operation unit 50 operates the frequency characteristic based on the LPC coefficient and the index of the adaptive code (operates so as to be close to the frequency characteristic of the input audio signal). ). The detailed configuration of the frequency characteristic operation unit 50 has the configuration shown in FIG. 3 described above. The fixed code whose frequency characteristics have been manipulated as described above is gain-controlled by the gain controller 49 and is provided to the adder 51.

The adder 51 adds the given adaptive code and fixed code, and supplies the added signal to the weighted synthesis filter 52 as an excitation signal. The weighted synthesis filter 52 convolves the excitation signal with LPC coefficients to obtain and output a synthesized speech signal.

The excitation signal output from the adder 51 is supplied to the adaptive codebook 43 again. At this time, the adaptive code book 43 updates the adaptive code using the excitation signal.

The code-excited linear predictive decoder performs the above-described processing every time a coded speech signal is supplied, and thus for each subframe.

Accordingly, the code-excited linear predictive decoder according to this embodiment has a configuration for processing a given LSP parameter, and has a pulse codebook 45 as a sound source.
And the frequency characteristic operation unit 50 that brings the frequency characteristics of the fixed sound source closer to the frequency characteristics of the input audio signal, thereby realizing the effect of the above-described code excitation linear prediction encoder.

(C) Other Embodiments The above embodiment relates to a forward-type code-excited linear prediction encoder and a decoder. The present invention relates to a backward-type code-excited linear prediction encoder. And a decoder.

Further, the above-mentioned embodiment is configured with consideration for the encoding speed of 4 bit / s or less, but it is needless to say that the present invention can be applied to an encoder and a decoder having a higher encoding speed. is there. If the coding speed allows, the statistical codebook and the pulsed codebook may not always be selected but both may always be operated effectively.

The frequency characteristic operating units 28 and 5 of the above embodiment
In the case of 0, both the frequency characteristic operation using the LPC coefficient and the frequency characteristic operation using the pitch lag are performed, but at least one of the operations may be performed.

[0064]

According to the code excited linear predictive encoder and decoder of the present invention, the frequency characteristic of the fixed code output from the fixed codebook, because the operation close to the frequency characteristics of the input speech signal, The sound source parameters correspond well to the actual sound source, and the quality of the reproduced sound can be improved from this point regardless of the coding speed .

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a code excitation linear prediction encoder.

FIG. 2 is a block diagram showing an embodiment of a code excitation linear prediction decoder.

FIG. 3 is a block diagram illustrating a detailed configuration of a frequency characteristic operation unit (50) according to the embodiment.

[Explanation of symbols]

3: multiplexing unit, 11: LSP parameter analysis unit, 12 ...
LSP parameter encoding unit, 13, 41 ... LSP parameter decoding unit, 14, 42 ... LPC coefficient conversion unit, 20,
43: Adaptive codebook, 21, 44: Statistical codebook, 22, 45: Pulse codebook, 25, 51 ...
Adder, 26, 48 ... fixed code selection switch, 28,
50: frequency characteristic operation unit, 29, 52: weighted synthesis filter, 30: subtractor, 31: error power sum calculation unit, 32
... Minimum error power code selector, 40 ... Demultiplexer.

────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Yoshihiro Ariyama 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) References JP-A-2-282800 (JP, A) JP-A-2-148926 (JP, A) JP-A-1-54497 (JP, A) JP-A-3-101800 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G10L 3 / 00-9/20 H03M 7/30 H04B 14/04 JICST file (JOIS)

Claims (8)

(57) [Claims] 1. A vocal tract parameter which is linearly predicted for an input speech signal.
Vocal tract parameter generating means for obtaining data, and quantizing the vocal tract parameters to obtain a quantized vocal tract parameter.
A vocal tract parameter quantizer that outputs at least parameters
Stage and an adaptive code representing the optimal code of the past input speech signal
And an adaptive code book that stores a predetermined statistical code and / or pulse code
The fixed code is stored based on the vocal tract parameters.
Frequency characteristics of the input audio signal determined at the time
Using the impulse response of the transfer function to be converted,
A frequency characteristic operation unit for performing convolution processing on the code, the adaptive code on which the convolution processing has been performed, and the fixed code;
And a gain controller for multiplying the gain by the gain and the gain, respectively, and the convolution multiplied by the gain by the gain controller.
Add the processed adaptive code and the fixed code
Excitation signal generating means for generating an excitation signal by using the quantized vocal tract parameters and the excitation signal.
A synthesis filter for creating a synthesized voice signal, and evaluating an error between the input voice signal and the synthesized voice signal.
The optimum time before the input audio signal at the current time
Select the adaptive code and fixed code, and corresponding index
A code selecting unit for determining a vocal tract parameter, the adaptive code and the fixed code.
Multiplexed with the index and output as an encoded audio signal.
Code excitation linear circuit having a multiplexing section
Predictive encoder. 2. The code-excited linear prediction code according to claim 1, wherein said impulse response is an impulse response of a transfer function H (z) represented by the following equation. Chemist. H (z) = 1 / (1−εz− ^L ) where ε is a constant satisfying 0 <ε ≦ 1, and L is a pitch lag calculated from the index of the adaptive code. 3. The code-excited linear prediction encoder according to claim 1, wherein the impulse response is an impulse response of a transfer function H (z) represented by the following equation. Chemist. H (z) = (1−Σγ ⁱ aiz− ⁱ ) / (1−Σv ⁱ aiz− ⁱ ) where ai (i is 1 to M) is a tap coefficient, and γ and ν are 0
<Γ <1 and 0 <ν <1. 4. The code excitation linear predictive encoder according to claim 1, wherein the impulse response is a transfer function H1 (z) represented by the following equation:
A code excitation linear predictive encoder characterized by using an impulse response in which the impulse response of the transfer function H2 (z) and the impulse response of the transfer function H2 (z) are cascaded. H1 (z) = (1−Σγ ⁱ aiz− ⁱ ) / (1−Σv ⁱ aiz− ⁱ ) H2 (z) = 1 / (1−εz− ^L ) where ai (i is 1 to M) Is the tap coefficient, and γ and ν are 0
<Γ <1 and 0 <ν <1 constants, ε is a constant satisfying 0 <ε ≦ 1, and L is a pitch lag calculated from the index of the adaptive code. 5. The method according to claim 1, wherein the code excitation linear prediction encoder supplies
Vocal tract parameters and adaptive code
Index and fixed code index.
Demultiplexing means, and a past input corresponding to the index of the adaptive code.
An adaptive code that outputs a speech signal code as an adaptive code.
And a predetermined book corresponding to the index of the fixed code.
A fixed code book that outputs a fixed code, and the fixed code is based on the vocal tract parameters.
Determined at the time selected by the code excitation linear prediction encoder
Of the transfer function to be converted into the frequency characteristic of the input audio signal
Using the impulse response, convolve the fixed code
A frequency characteristic operation unit for performing processing, the adaptive code subjected to the convolution processing, and the fixed code.
And a gain controller for multiplying the gain by the gain and the gain, respectively, and the convolution multiplied by the gain by the gain controller.
Add the processed adaptive code and the fixed code
Excitation signal generation means for generating an excitation signal by using the vocal tract parameter and the excitation signal.
And a synthesis filter for generating a signal.
Code-excited linear predictive decoder. 6. The code-excited linear prediction decoder according to claim 5, wherein the impulse response is an impulse response of a transfer function H (z) represented by the following equation. Chemist. H (z) = 1 / (1−εz− ^L ) where ε is a constant satisfying 0 <ε ≦ 1, and L is a pitch lag calculated from the index of the adaptive code. 7. The code-excited linear prediction decoder according to claim 5, wherein the impulse response is an impulse response of a transfer function H (z) represented by the following equation. Chemist. H (z) = (1−Σγ ⁱ aiz− ⁱ ) / (1−Σv ⁱ aiZ− ⁱ ) where ai (i is 1 to M) is a tap coefficient, and γ and ν are 0
<Γ <1 and 0 <ν <1. 8. The code-excited linear prediction decoder according to claim 5, wherein the impulse response is a transfer function H1 (z) represented by the following equation:
A code-excited linear predictive decoder characterized by using an impulse response obtained by cascading an impulse response of the transfer function H2 (z) and an impulse response of the transfer function H2 (z). H1 (z) = (1−Σγ ⁱ aiz− ⁱ ) / (1−Σv ⁱ aiz− ⁱ ) H2 (z) = 1 / (1−εz− ^L ) where ai (i is 1 to M) Is the tap coefficient, and γ and ν are 0
<Γ <1 and 0 <ν <1 constants, ε is a constant satisfying 0 <ε ≦ 1, and L is a pitch lag calculated from the index of the adaptive code.