JPH09269799A - Voice coding circuit provided with noise suppression function - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the quality of a coded voice signal by performing a noise suppression without exerting adverse influence on a voice coding processing. SOLUTION: In the voice coding circuit, the power level of a background noise to be included in a transmitted input signal is estimated by a noise suppression control part 20 and a modifying quantity (r) is calculated from the estimated value and the power of the transmitted input signal. Then, the code word of the frame power calculated in the frame power energy calculating part 31 of a voice coding part 30 and the code word of the zero input response signal calculated in a zero input response calculating part 36 and code words of gain values GS, PO of the long-term productive rag and the code respectively selected in a long-term predictive rag selecting part 38 and a code book searching part 39 are respectively modifyingly operated based on the modifying quantity (r) and then code words of respective coding parameters after the modifying operations are transmitted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding circuit provided in a digital voice communication device such as a digital automobile / portable telephone device, a digital cordless telephone, a digital wire telephone device and the like.

[0002]

2. Description of the Related Art In communication devices such as digital mobile phone devices, CELP (Code Excited Linear Predictio) is generally used.
n) Low bit rate audio coding method such as the method is used. By using this type of coding method, it is possible to make a good voice call even in an environment where the background noise is relatively large. For details on the CELP method, please refer to “Code-Excited” by MR Schroeder and BSAtal.
Linear Prediction (CELP): High-Quality Speech At Ve
ry Low Bit Rates ”in Proc. ICASSP, 1985.pp.937-939.

However, in a high noise environment such as in a bus or a commuter train, background noise remarkably reduces the clarity of speech, so various researches have been made on a noise canceller that removes noise and provides only speech for coding. ing. One of them is, for example, the "noise reduction processing method" disclosed in Japanese Patent Laid-Open No. 7-193548.

FIG. 3 shows an outline of a speech coding circuit having a noise reduction processing function shown in this example. In the figure, the transmission signal output from the microphone 1 is converted into a digital signal by an analog / digital (A / D) converter 2 and then input to a frame power calculation unit 3, where the power for each frame is calculated. Calculated. Further, the level discrimination circuit 7 determines the threshold value of the noise suppression amount based on the level of noise mixed in the transmission signal.
This threshold value is given to the suppression ratio calculation circuit 4. The suppression ratio calculation circuit 4 calculates the suppression ratio for noise based on the power of the transmission signal calculated by the frame power calculation unit 3 and the threshold value of the noise suppression amount determined by the level discrimination circuit 7. To be done. This suppression ratio is smoothed by the smoothing circuit 5 to suppress variations for each frame, and then given to the noise reduce circuit 6.
The noise reduce circuit 6 multiplies the transmission signal by the suppression ratio, thereby suppressing the noise component in the transmission signal. Then, the transmission signal in which this noise is suppressed is supplied to the voice encoding unit 8 and subjected to voice encoding processing.

That is, the conventional speech encoding circuit calculates a suppression ratio according to the noise level, suppresses the noise of the transmission input signal based on this suppression ratio, and then the signal is encoded by the speech encoding unit 8. It is configured to input to and perform voice encoding.

[0006]

However, such a conventional speech coding circuit has the following problems to be solved. That is, generally, a speech encoding circuit using the CELP method or the like refers to signals before and after a frame to be encoded, performs LPC analysis and pitch analysis, and encodes and transmits the LPC coefficient and pitch period obtained by this. Is configured to. For this reason, when the speech input signal is noise-suppressed and then speech-encoded like the above-described conventional speech encoding circuit, the signal value at the frame boundary is distorted, and the analysis results of the LPC analysis and the pitch analysis are not obtained. It could be accurate.

The present invention has been made in view of the above circumstances, and an object thereof is to enable noise suppression processing without adversely affecting speech coding processing,
Accordingly, it is an object of the present invention to provide a speech coding circuit having a noise suppression processing function that can improve the quality of a coded speech signal.

[0008]

In order to achieve the above object, the present invention analyzes a transmission input signal framed by a certain length for each frame and generates a codeword of a predetermined coding parameter. The audio coding processing means, the noise level estimating means for estimating the power of noise mixed in the transmission input signal for each frame of the transmission input signal, and the code word changing means. Then, by the code word changing means, the coding generated by the speech coding processing means based on the difference between the noise power estimation result by the noise level estimating means and the power of the corresponding transmission input signal frame. Among the codewords of the parameters, at least the codeword representing the frame power is changed.

Therefore, according to the present invention, the noise suppression processing is performed on the code word representing the frame power generated by being analyzed by the voice encoding processing means. That is, noise cancellation processing is performed on the encoded signal. Therefore, the speech input signal before noise suppression processing can be supplied to the speech coding processing means, whereby the speech coding unit 30 is not affected by the noise cancellation processing at all, and LPC analysis, pitch analysis, etc. are performed. It is possible to accurately perform the interframe coding processing of. Also,
Since noise suppression processing can be performed by changing the code word of a specific encoding parameter, noise suppression processing can be performed with a smaller amount of calculation than when performing arithmetic processing for noise suppression on a transmission input signal before encoding. Can be performed.

According to the present invention, the voice encoding processing means is
When a code is selected from the adaptive codebook or the noise codebook and the driving signal is coded using the selected code and the gain of the code, the codeword changing means has the gain of the code. It is characterized in that the code word representing is changed based on the difference between the noise power estimation result by the noise level estimating means and the power of the corresponding transmission input signal frame.

Further, according to the present invention, when the speech coding processing means has a means for performing a predetermined weighting for each frame of the transmission input signal and subtracting the zero input response component from the weighted signal. In the code word changing means, the code word representing the power of the zero input response component or the code word representing the power of the signal after being weighted by the weighting filter is used as the noise power estimation result by the noise level estimating means. , Is also characterized in that it is changed based on the difference from the power of the corresponding transmission input signal frame.

As described above, in addition to the process of changing the code word representing the frame power, the code word representing the code gain,
By changing the codeword representing the power of the zero input response component or the codeword representing the power of the signal after being weighted by the weighting filter,
It is possible to improve the consistency with the above-described change processing for the frame power code word, and as a result, it is possible to perform more accurate noise cancellation processing as compared with the case where only the change processing for the frame power code word is performed. Become.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION

(First Embodiment) In this embodiment, the present invention is applied to a full-rate voice encoding circuit, and FIG. 1 shows a functional configuration of a voice encoding circuit having a noise suppressing function according to this embodiment. It is a block diagram.

The speech coding circuit of this embodiment comprises a frame division section 10, a noise suppression control section 20, a speech coding section 30, and a multiplexing section (MUX) 40. The frame division unit 10 divides the transmission input signal, which is output from a microphone (not shown) and then digitized by the A / D converter, into a predetermined frame length. Is the noise suppression control unit 2
0 and the speech coding unit 30 respectively. The frame length is, for example, 16 in the A / D converter.
It is set to correspond to 0 sample.

The noise suppression control section 20 comprises a noise level estimation section 21 and a change amount calculation section 22. The noise level estimation unit 21 estimates the noise level for each frame of the transmission input signal. The noise level is estimated by extracting only the background noise mixed in the transmission input signal, obtaining the frame average value of the power, and calculating the logarithmic value of the frame average value. There are various methods for estimating the noise level, but the method described in Japanese Patent Application No. 7-52137, "Noise Level Estimating Circuit" is suitable.

In the change amount calculation unit 22, based on the noise level estimation value calculated by the noise level estimation unit 21, the change amount r used for changing the codeword in the speech coding unit 30 described later is calculated. Calculation is performed. The calculation of the change amount r is performed as follows, for example.

[0017]

[Equation 1]

N is a noise level, L, A and B are constants, and these constants L, A and B are, for example, L = 8 and A =
16, B = 12 is set. On the other hand, the voice encoding unit 30
Is configured as follows. That is, the transmission input signal output from the frame division unit 10 is transmitted to the voice encoding unit 3
Within 0, the frame energy calculator 31, LPC
It is input to the coefficient calculation unit 32 and the subframe division unit 33, respectively.

First of all, the frame energy calculation unit 3
In 1, the logarithmic value of the root mean square value of the energy is calculated for each frame of the transmission input signal, and the calculated value is vector-quantized and then encoded. The calculation method of this energy is shown below. That is, the frame energy R (0) is calculated by the following equation.

[0020]

[Equation 2]

Next, R (0) is converted as shown in the following equation with reference to full scale. In addition, full scale Rma
x is defined as the square of the maximum sample amplitude. RdB = 10 log ₁₀ (R (0) / Rmax) Then, RdB is quantized into 32 levels. The 32 quantized value of this RdB takes a value from the minimum value -66 (corresponding to the 0 code of R0) to the maximum value -4 (corresponding to the 31 code of R0). The quantization step size is
2 (2 dB step). R0 is a value that minimizes the following equation. abs (R0- (RdB + 66) / 2) However, R0 takes an integer value from 0 to 31 corresponding to 32 code words of R0.

A quantized value Rq (0) of R0 is given by the following equation. Rq (0) = Rmax (10 ^{(2R0 -66) / 10} ) The above frame energy calculation method is described in Chapter 5 5 of Digital Car Telephone System Standard (RCRSTD-27C) issued by Radio System Development Center. ．
See Section 1.1.5 for details.

Further, in the frame energy calculation unit 31,
The code amount of the frame energy of the transmission input signal calculated above is used as the change amount r calculated by the change amount calculation unit 22.
The calculation for changing is performed using. That is,
The value of R (0) is converted into R (0) × r 2.

The LPC coefficient calculator 32 obtains the autocorrelation matrix for each frame of the transmission input signal, obtains the LPC coefficients up to the 10th order by the FLAT algorithm, and encodes them.

The subframe division unit 33 divides the frame of the transmission input signal into four subframes and inputs these subframes to the weighting filter 34. The weighting filter 34 is composed of a perceptual noise weighting filter, and the filter coefficient is set according to the LPC coefficient obtained by the LPC coefficient calculating section 32. Then, the sub-frame of the transmission input signal is passed through this filter to generate an excitation signal weighted in consideration of hearing. The construction and characteristics of this weighting filter are described in detail in Chapter 5, Section 5.1.1.1.1 of RCR STD-27C.

The excitation signal output from the weighting filter 34 is input to the subtractor 35, which subtracts the zero input response signal of the previous frame. This zero input response signal is created by the zero input response calculation unit 36.

The zero-input response calculation section 36 determines a zero-input response signal, which is the influence of the previous frame on the current frame, on the basis of the LPC coefficient and the excitation signal of the previous frame output from the local decoder 41 described later. create. Further, when creating the zero-input response signal, the zero-input response calculation unit 36 uses the zero-input response calculation unit 36 in order to maintain consistency with the frame energy codeword changing process performed in the frame energy calculation unit 31. The signal level is multiplied by 1 / r according to the change amount r obtained by the change amount calculation unit 22. That is, the zero input response calculation unit 36
The numerator of the synthesis filter H (z) shown in the following equation is r ⁻¹ .

[0028]

(Equation 3)

The excitation signal from which the zero input response signal of the previous frame is subtracted in the subtractor 35 is the code selection unit 3
7, input to the long-term prediction lag selection unit 38 and the codebook search unit 39, respectively.

Of these, the long-term prediction lag selecting section 38 first performs a process for approximating the excitation signal of the current subframe by using a part of the excitation signal of the past subframe. That is, since an audio signal usually has periodicity, an approximate signal, that is, a long-term predicted vector is created by repeating a plurality of past samples. This period is called a lag and is encoded.

The codebook search unit 39 has a noise codebook. Then, the long-term predicted lag selection unit 3
In order to make the long-term predicted vector approximated by 8 more accurate, an appropriate noise code is selected from the above noise codebook. This selected noise code is called a code vector and this code is coded. For details of the search processing performed by the codebook search unit 39, refer to Chapter 5, 5.1. Of RCRSTD-27C.
1.6.3. Section and 5.1.1.1.6.4. Described in the section.

Further, the long-term prediction vector and the code vector created in the long-term prediction lag selecting section 38 and the codebook search section 39 are mutually given a long-term prediction coefficient β and a scalar gain γ of the code vector, respectively. Are added, which results in the excitation function in the current subframe. This excitation function ex (n) is expressed as follows. ex (n) = βc ₀ (n) + γc ₁ (n) 0 ≦ n ≦ N−1, where c ₀ : non-weighted long-term prediction vector c ₁ : non-weighted code vector selected from codebook N = 40 = It is 160/4. The energy in each excitation vector is given by the following equation.

[0033]

(Equation 4) Let RS be the approximate residual energy in a given subframe. RS is a function of N, R'q (0) and the normal prediction gain of the LPC filter.

[0034]

(Equation 5) Where r _{i is} the i-th reflection coefficient (i = 1 to 1) of the sub-frame corresponding to the set of direct filter coefficients α _i of the sub-frame
0). The long-term prediction coefficient β and the scalar gain γ of the code vector
Are converted into two new parameters GS and P0 respectively. GS is an energy offset parameter for adjusting the estimated value of RS, and is defined as R = GS · RS. The P0 is the ratio of energy generated from the long-term prediction vector as a percentage of the total excitation energy in a sub-frame is defined as ^{P0 = β 2 Rz (0)} / R 0 ≦ P0 ≦ 1. Therefore, β and γ are rewritten into the parameters GS and P0 by the following equations, respectively.

[0035]

(Equation 6) The code selection unit 37 vector-quantizes each of the parameters GS and P0 and then codes them. That is, when β and γ described above are replaced with the equivalent formula by Rz (k),

[0036]

(Equation 7) Note that Rpp, Rpc, and Rcc are represented as follows.

[0037]

(Equation 8)

The GS and P0 are quantized. {GS,
The first step in quantizing the P0} vector is the above E
It consists of the calculation of the parameters required by the formula. Rcc (k, j) k = 0,1 j = k, 1 Rz (0), Rz (1) RS Rpc (k) k = 0,1 a, b, c, d, e The expression is evaluated for each vector in the {GS, P0} codebook, choosing the vector that minimizes the weighted error.

Further, the code selecting unit 37 performs a process for changing the code word of the parameters GS and P0 based on the change amount r calculated by the noise suppression control unit 20. That is, the values of c, d, and e are replaced with c / r, d / r, and e / r, respectively.

Each coding parameter created as described above, that is, the frame energy code word R (0) × r calculated by the frame energy calculation unit 31 and further subjected to the change processing for noise suppression. And the LPC coefficient calculated by the LPC coefficient calculation unit 32, and the long-term prediction lag selection unit 3
8 and the codebook search unit 39 respectively, the long-term prediction vector and the code vector, and the code word of the parameters GS and P0 calculated by the code selection unit 37 and subjected to the code word changing process for low noise suppression. Are multiplexed by the multiplexing unit 40 and sent from the transmission circuit unit (not shown) to the receiving side device.

At the same time, each of the above coding parameters is input to the local decoder 41. The local decoder 41 generates an excitation signal that is predicted to be decoded and reproduced by the speech decoding unit of the reception side device, based on the above-mentioned encoding parameters. Then, the generated excitation signal of the previous subframe is used in the zero input response calculation unit 36 and the long-term prediction lag selection unit 38 for the calculation of the zero input response and the calculation of the long-term prediction coefficient, respectively.

With such a configuration, when the transmission signal is input, the noise suppression control unit 20 estimates the power of the background noise included in the transmission input signal, and the power of the noise is estimated according to the power value of the noise. Then, the change amount r for noise suppression is calculated.

Further, the transmission input signal is directly input to the voice encoding unit 30 without being subjected to noise suppression processing. Then, in each calculation unit of the voice encoding unit 30, each encoding parameter for encoding and transmitting the transmission input signal is calculated. That is, the frame energy calculation unit 31 calculates the code word R (0) of the frame energy, and the LPC coefficient calculation unit 32 calculates the LPC coefficient. The long-term prediction lag selection unit 38 and the codebook search unit 39 calculate long-term prediction vectors and code vectors, respectively, and the code selection unit 37 calculates the code words of the parameters GS and P0.

At this time, in the frame energy calculating section 31, the code word R of the frame energy is calculated based on the change amount r calculated by the noise suppression control section 20.
Change processing for noise suppression is performed on (0). That is, the noise canceling process is performed not on the transmission input signal before being input to the voice encoding unit 30, but on the code word of the frame energy obtained by the voice encoding unit 30.

Further, in the zero input response calculation unit 36 and the code selection unit 37, in order to match the change of the frame energy with the code word, the change amount r calculated by the noise suppression control unit 20 is used as a basis. , The zero input response signal and the code word of the parameters GS and P0 are changed.

The respective coding parameters calculated by the voice coding unit 30 are multiplexed by the multiplexing unit 40 and transmitted from the transmission circuit unit to the communication partner device. As described above, in this embodiment, in the full-rate speech encoding circuit, the noise suppression control unit 20 estimates the power level of the background noise included in the transmission input signal, and the estimated value and the transmission input signal are calculated. Of the frame energy calculated by the frame energy calculation unit 31 of the speech encoding unit 30 and the zero input response calculation unit 36 based on the change amount r. The obtained codeword of the zero input response signal and the codewords of the coding parameters GS and P0 corresponding to the long-term prediction lag and noise code selected by the long-term prediction lag selection unit 38 and the codebook search unit 39, respectively. A change operation is performed, and the codeword of each coding parameter after this change operation is transmitted.

Therefore, according to this embodiment, the noise canceling process is performed for each coding parameter after coding in the speech coding unit 30, and therefore the speech coding unit 30 is subjected to the noise cancellation process before the noise cancellation. A speech input signal can be input, which allows the speech encoding unit 30 to accurately calculate the LPC coefficient without being affected by the noise canceling process. Therefore, the quality of coded speech can be improved.

Further, since the operation for noise cancellation can be performed by operating the code word of the encoding parameter, the amount of operation can be increased as compared with the case where the operation for noise cancellation is performed on the transmission input signal. It is possible to reduce the processing delay amount of the speech coding processing including noise cancellation, and to speed up the coding.

In the above embodiment, the codeword changing process is performed to match the codeword of the zero input response signal and the codeword of the coding parameters GS and P0 with the change of the frame energy codeword. However, the code word changing process for these matching does not necessarily have to be performed.

(Second Embodiment) In this embodiment, the present invention is applied to a half-rate speech coding circuit. FIG. 2 shows a speech coding circuit having a noise suppressing function according to this embodiment. It is a block diagram showing functional composition. In the figure, the same function parts as those in FIG.

The speech coding section 50 is provided with a pitch analysis section 54 and a filter 55 in addition to the frame energy calculation section 51, the LPC coefficient calculation section 52 and the subframe division section 53.

The frame energy calculator 51 obtains the average amplitudes amp _n (n = 0, 1, 2, 3) of the four subframes constituting one frame of the transmission input signal, and these amplitudes amp The logarithmic value of the root mean square value of _n is vector-quantized and then coded. The calculation formula is shown below. That is, first, the average amplitude am of the subframes
p _n is calculated as follows.

[0053]

[Equation 9] This amplitude amp _n is μ-law converted to μ shown in the following equation.
It becomes amp _n .

[0054]

(Equation 10)

However, μ = 100, Smax is the input PCM
It is the maximum amplitude of the signal. Next, i that minimizes err in the following equation is set as the power parameter index POW.

[0056]

[Equation 11] Then, the quantized power Pn is obtained as follows.

[0057]

(Equation 12)

The quantized power Pn obtained by the frame energy calculator 51 is input to the noise suppression controller 60. The noise suppression control unit 60 includes a noise level estimation unit 61 and a change amount calculation unit 62. The noise level estimation unit 61 estimates the logarithmic value of the subframe average value of the background noise power included in the transmission input signal based on the quantized power Pn obtained by the frame energy calculation unit 51.

In the change amount calculation unit 62, the change amount r for changing the code word of the subframe energy based on the noise power estimation value calculated by the noise level estimation unit 61.
Is calculated, and the calculated change amount r is supplied to the frame energy calculation unit 51.

In the frame energy calculation section 51, the amplitude amp _n is calculated in the frame energy calculation process.
Is calculated using the change amount r obtained by the change amount calculation unit 62. That is, the value of the # 038 _n are converted to # 038 _n × r. Then, the converted amp _n × r is vector-quantized and coded.

The LPC coefficient calculation unit 52 obtains the LPC coefficients up to the 10th order for each subframe of the transmission input signal, converts the LPC of the odd frame into the LSP, and encodes the LSP. Further, the filter 55 calculates a filter coefficient used for weighting the perceptual sensation. The pitch analysis unit 54 analyzes the pitch period of the transmission input signal frame output from the frame division unit 10, and thereby calculates the filter coefficient used by the filter 55.

The subframe division unit 53 divides the current frame consisting of 320 samples into four subframes. Then, the divided subframe is filtered by the filter 5
Enter in 5. The filter 55 includes the subframe division unit 5
For the sub-frame of the transmission input signal output from 3,
Filtering processing for perceptual weighting, spectrum weighting, and pitch weighting is performed to create a reference signal.

The details of the processing in the LPC coefficient calculation section 52, the pitch analysis section 54 and the filter 55 are as follows.
5.2.1.5 of RCRSTD-27C Chapter 5 respectively
Sections 5.2, 1.7 and 5.2.1.8.

The reference signal output from the filter 55 is input to the multiplier 56. In the multiplier 56, the amplitude am performed in the frame energy calculator 51 is calculated.
The level of the reference signal output from the filter 55 is multiplied by r using the change amount r obtained by the change amount calculation unit 62 in order to achieve consistency with the _pn codeword changing process.

The reference signal output from the multiplier 56 is subsequently input to the subtractor 57, where the zero input response signal of the previous subframe created by the zero input response calculation section 58 is subtracted. Then, the reference signal output from the subtractor 57 is the code selection unit 63, the first codebook
Search unit 59 and second codebook search unit 65
Respectively.

The first codebook search unit 59 is an adaptive codebook (ACB) or a fixed codebook (FCB) created from past frames in order to approximate the reference signal output from the subtractor 57. Is searched to select one of the codes, and the selected code is encoded. The second codebook search unit 65 searches the noise codebook (SCB) to select an appropriate code and encodes the selected code in order to improve the approximation accuracy of the reference signal.

The code selecting section 63 weights the combined signal SYNp of the signals selected by the first codebook search section 59 and the weighted combined signal SYNSC of the signals selected by the second codebook search section 65. To each of them, the gain is given to each of them, and the sum is obtained, thereby becoming an approximate signal of the above reference signal. The gain is normalized and coded by the power of the combined signal SYNp or the combined signal SYNSC and the estimated residual power RS of the current subframe. The estimated residual power RS of the current subframe is estimated using the frame power and the gain of the filter 55.

The details of the codebook search process performed in the first and second codebook search units 59 and 65 and the coding process of the subframe residual gain in the code selection unit 63 are described in RCRSTD-27.
C Chapter 5, section 5.2.1.8.

In this way, each coding parameter created by the speech coding unit 50, that is, the code word of the subframe energy calculated by the frame energy calculation unit 51 and further subjected to the change processing for noise suppression. amp _n × r
And the LPC coefficient calculated by the LPC coefficient calculation unit 52,
The signals SYNp and SYNSC selected by the first and second codebook search units 59 and 65, the coding gain obtained by the code selection unit 63, and the transmission circuit unit (not shown) multiplexed by the multiplexing unit 40. Is sent to the receiving side device.

At the same time, the above encoding parameters are input to the local decoder 66. In the local decoder 66, an excitation signal that is predicted to be decoded and reproduced by the speech decoding unit of the reception side device is generated based on each of the above coding parameters. Then, the generated excitation signal of the previous subframe is used in the zero input response calculation unit 58 and the first codebook search unit 59 for calculating the zero input response and ACB / FCB search, respectively. To be done.

With such a configuration, as in the case of using the full-rate speech coding unit 30 described in the first embodiment, the frame energy calculation unit 51 is used.
In the above, based on the change amount r calculated by the noise suppression control unit 60, the change process for noise suppression is performed on the amplitude amp _n of the root-mean-square value of the subframe, and the amplitude amp _n × _n after the change process is performed. r is vector quantized and encoded. That is, the noise canceling process is performed not on the transmission input signal before being input to the voice encoding unit 50 but on the code word of the subframe energy obtained by the voice encoding unit 50.

Further, in the multiplier 56, the filter 5
The reference signal output from the signal No. 5 is subjected to change processing based on the change amount r calculated by the noise suppression control unit 60, thereby changing to the code word of the code gain obtained by the code selection unit 63. Processing is performed to match the subframe energy codeword changes.

As described above, according to this embodiment, in the half-rate voice encoding circuit, the noise suppression control unit 6
Based on the change amount r obtained by 0, the speech encoding unit 50
The process of changing the amplitude amp _n of the root mean square value of the subframe calculated by the frame energy calculation unit 51 and the process of changing the reference signal generated by the filter 55 are performed. Therefore, also in this embodiment,
The noise canceling process is performed on each coding parameter after coding in the speech coding unit 50. Therefore, the speech input signal before noise cancellation can be input to the speech coding unit 50, As a result, the speech encoding unit 50 can accurately perform the LPC analysis and the pitch analysis without being affected by the noise canceling process. Therefore, the quality of coded speech can be improved.

The present invention is not limited to the above embodiments. For example, in each of the above embodiments, the CEL
The description has been given by taking the P-type voice encoding circuit as an example.
The present invention may be applied to a voice encoding circuit that adopts another vector quantization method such as the C method.

In addition, with respect to the configurations of the voice coding unit and the noise suppression control unit, the coding parameters to be changed, the code word change processing method, and the like, various modifications can be made without departing from the scope of the present invention. There is no end.

[0076]

As described above in detail, according to the present invention, the speech coding processing means for analyzing the framed transmission input signal for each frame to generate the code word of the predetermined coding parameter, and the transmission input. A noise level estimating means for estimating the power of noise mixed in the transmission input signal for each frame of the signal and a code word changing means are provided, and the code word changing means causes noise by the noise level estimating means. Based on the difference between the power estimation result and the power of the corresponding transmission input signal frame, at least the codeword representing the frame power is changed among the codewords of the coding parameters generated by the speech coding processing means. I have to.

Therefore, according to the present invention, the noise suppression process can be performed on the code word representing the frame power generated by being analyzed by the voice encoding processing means.
Therefore, the speech input signal before noise suppression processing can be supplied to the speech coding processing means, whereby the speech coding unit 30 is not affected by the noise cancellation processing at all, and LPC analysis, pitch analysis, etc. are performed. It is possible to accurately perform the interframe coding processing of.

That is, according to the present invention, the noise suppression processing can be performed without adversely affecting the speech encoding processing, whereby the quality of the encoded speech signal can be further improved. A voice encoding circuit having a function can be provided.

[Brief description of drawings]

FIG. 1 is a block diagram showing a functional configuration of a voice encoding circuit having a noise suppressing function according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a functional configuration of a voice encoding circuit having a noise suppressing function according to a second embodiment of the present invention.

FIG. 3 is a diagram showing a schematic configuration of a conventional speech encoding circuit with a noise reduction function.

[Explanation of symbols]

 10 ... Frame division unit 20, 60 ... Noise suppression control unit 21, 61 ... Noise level estimation unit 22, 62 ... Change amount calculation unit 30, 50 ... Speech coding unit 31, 51 ... Frame energy calculation unit 32, 52 ... LPC coefficient calculation unit 33, 53 ... Subframe division unit 34, 55 ... Weighting filter 35, 57 ... Subtractor for subtracting zero input response 36, 58 ... Zero input response calculation unit 37, 63 ... Sign selection unit 38 ... Long-term prediction Lag selection unit 39, 59, 65 ... Codebook search unit 40 ... Multiplexing unit (MUX) 41, 66 ... Local decoder 54 ... Pitch analysis unit 56 ... Multiplier for changing codeword 59, 61 ... Code selection unit

Claims (4)

[Claims] 1. A voice encoding processing unit for analyzing a frame-by-frame transmission input signal for each fixed length to generate a code word of a predetermined encoding parameter, and for each frame of the transmission input signal. , A noise level estimation means for estimating the power of noise mixed in the transmission input signal, and a difference between the noise power estimation result by the noise level estimation means and the power of the corresponding transmission input signal frame. And a code word changing means for changing the content of at least the code word representing the power among the code words of the coding parameter generated by the speech coding processing means. Speech coding circuit equipped with. 2. The speech coding processing means has means for selecting a code from an adaptive codebook or a noise codebook and coding a drive signal using the selected code and the gain of the code. The code word changing means changes the code word representing the gain of the code based on the difference between the noise power estimation result by the noise level estimating means and the power of the corresponding transmission input signal frame. A speech coding circuit having a noise suppression processing function according to claim 1. 3. A code word when the speech coding processing means has a means for performing a predetermined weighting for each frame of a transmission input signal and subtracting a zero input response component from the weighted signal. The changing means changes the codeword representing the power of the zero input response component based on the difference between the noise power estimation result by the noise level estimating means and the power of the corresponding transmission input signal frame. A speech coding circuit having a noise suppression processing function according to claim 1. 4. The speech coding processing means includes a means for performing a predetermined weighting by a weighting filter for each frame of a transmission input signal and subtracting a zero input response component from the weighted signal, The code word changing means determines the code word representing the power of the signal after being weighted by the weighting filter as a difference between the noise power estimation result by the noise level estimating means and the power of the corresponding transmission input signal frame. The speech coding circuit having the noise suppression processing function according to claim 1, wherein the speech coding circuit is modified based on the above.