CN101388215A - A method and device for noise shaping - Google Patents

Abstract

The present invention provides a method and device for noise shaping, which performs adaptive noise shaping processing on speech signal s(n) to obtain signal s ₁ (n); performs fixed noise shaping on said signal s ₁ (n) Processing to obtain the signal u(n); voice encoding is performed on the signal u(n) to form an output signal at the encoding end. A noise shaping method and device are also provided, wherein the speech signal s(n) is subjected to fixed noise shaping processing to obtain a signal s ₁ ′(n); adaptive noise shaping is performed on the signal s ₁ ′(n) Reshaping processing to obtain signal u'(n); voice encoding is performed on said signal u'(n) to form an output signal of the encoding end. Since both the fixed noise shaping process and the adaptive noise shaping process are noise shaping processes on the speech signal itself, and do not involve the speech signal and the residual signal of the prediction filter, there is no need to make any changes at the decoding end The noise-shaped speech signal can be decoded and restored.

Description

A method and device for noise shaping

technical field

The invention relates to signal processing technology, in particular to a noise shaping method and device.

Background technique

Noise shaping is a technology that uses the principle of human auditory masking to shape the quantized noise spectrum generated by the codec, that is, adjusts the noise spectrum in the signal to approximate the shape of the speech spectrum, and uses the human ear masking effect to make the signal The noise in is not easy to detect. Usually, in the transmission of voice signals, in order to reduce the bandwidth occupied by the voice signal transmission, the voice signal is generally encoded and then transmitted, and then decoded at the receiving end to obtain the voice signal. In order to improve the subjective auditory quality of the decoded voice signal, The noise shaping process is often performed at the encoding end, and then the noise shaping process and the encoded speech signal are sent out.

Fig. 1 is the composition structure diagram of the noise shaping equipment in the prior art, as shown in Fig. 1, the processing process at the encoding end is: first calculate the error of the speech signal and its prediction signal obtained after passing through the prediction filter, and obtain the prediction residual signal d(n); d(n) is combined with the feedback signal fq(n) to obtain the input signal u(n) of the quantizer; u(n) is encoded by the quantizer to obtain the encoded signal uq(n) ; When performing feedback, the difference between the input signal u(n) of the quantizer and the output signal uq(n) of the quantizer is calculated to obtain the difference signal q(n); the difference signal q(n) is filtered by noise shaping The feedback signal fq(n) is obtained after shaping and filtering by the filter. The output signal uq(n) encoded by the quantizer reaches the decoding end after transmission, and the processing process at the decoding end is as follows: since the signal uq(n) output at the encoding end is essentially a speech signal and the prediction signal obtained after passing through the prediction filter The error is a residual signal obtained after processing. Therefore, at the decoding end, it is necessary to add the received signal uq(n) to the predicted signal obtained by the predictive filter to obtain the final restored speech signal. The restored speech signal is the speech signal after noise shaping.

However, in this noise shaping method in the prior art, the output signal at the encoding end is not the encoded signal of the speech signal after noise shaping, but the subtraction of the speech signal and the signal processed by the predictive filter to obtain The coded signal after the residual signal is processed by noise shaping and filtering, so the prediction filter must be added at the decoding end to compensate the residual signal to restore the speech signal, that is to say, when using this noise shaping method, at the same time It is also necessary to modify the original decoder.

Contents of the invention

Embodiments of the present invention provide a noise shaping method and device, so that the original decoding end does not need to be modified when performing noise shaping processing on speech signals.

A method of noise shaping, the method comprising:

Perform adaptive noise shaping processing on the speech signal s(n) to obtain the signal s ₁ (n);

performing fixed noise shaping processing on the signal s ₁ (n) to obtain a signal u(n);

Speech coding is performed on the signal u(n) to form an output signal at the coding end.

A method of noise shaping, the method comprising:

Perform fixed noise shaping processing on the speech signal s(n) to obtain the signal s ₁ '(n);

performing adaptive noise shaping processing on the signal s ₁ '(n) to obtain the signal u'(n);

Speech coding is performed on the signal u'(n) to form an output signal at the coding end.

A noise shaping device, the device comprising:

An adaptive noise shaping unit is used to perform adaptive noise shaping processing on the speech signal s(n) to obtain a signal s ₁ (n);

A fixed noise shaping unit, configured to perform fixed noise shaping processing on the signal s ₁ (n) obtained after the adaptive noise shaping unit performs adaptive noise shaping processing, to obtain a signal u(n);

The speech encoding unit is configured to perform speech encoding on the signal u(n) obtained after processing by the fixed noise shaping unit, to obtain an output signal at the encoding output terminal.

A noise shaping device, the device comprising:

The fixed noise shaping unit is used to perform fixed noise shaping processing on the speech signal s(n) to obtain the signal s ₁ '(n);

An adaptive noise shaping unit, configured to perform adaptive noise shaping processing on the signal s ₁ '(n) obtained by the fixed noise shaping unit to obtain a signal u'(n);

The speech encoding unit is configured to perform speech encoding on the signal u′(n) obtained by the adaptive noise shaping unit, to obtain an output signal of the encoding end.

It can be seen from the above technical solutions that in the method and device provided by the embodiments of the present invention, the voice signal is first subjected to adaptive noise shaping processing, and then fixed noise shaping processing is performed, and the signal after fixed noise shaping processing is performed Speech coding to form the output signal of the coding end; or, the speech signal is firstly subjected to fixed noise shaping processing, and then adaptive noise shaping processing is performed, and speech coding is performed on the signal after adaptive noise shaping processing to form the output signal of the coding end . Since both the fixed noise shaping process and the adaptive noise shaping process are noise shaping processes on the speech signal itself, and do not involve the speech signal and the residual signal of the prediction filter, there is no need to make any changes at the decoding end The noise-shaped speech signal can be decoded and restored.

Description of drawings

FIG. 1 is a structural diagram of a noise shaping device in the prior art;

FIG. 2 is a structural diagram of the first type of noise shaping device provided by an embodiment of the present invention;

Fig. 3 is a second composition structure diagram of the noise shaping device provided by the embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a specific noise shaping device provided by an embodiment of the present invention;

Fig. 5 is a schematic structural diagram of another specific noise shaping device provided by an embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

The noise shaping method provided by the embodiment of the present invention includes: performing fixed noise shaping on the speech signal s(n) and performing adaptive noise shaping processing according to the speech signal s(n) to form a signal u(n); The signal u(n) is speech coded to form an output signal at the coding end.

Wherein, the sequence of performing fixed noise shaping on the speech signal s(n) and performing adaptive noise shaping processing according to the speech signal s(n) may be to perform fixed noise shaping processing first, and then according to the speech signal s(n) n) performing adaptive noise shaping processing; or first performing adaptive noise shaping processing according to the speech signal s(n), and then performing fixed noise shaping processing.

The process of first performing adaptive noise shaping processing according to the speech signal s(n), and then performing fixed noise shaping processing may include the following processes:

Combine the voice signal s(n) with the second feedback signal f ₂ q ₂ (n) to obtain the signal s ₁ (n), and combine the signal s ₁ (n) with the first feedback signal f ₁ q ₁ (n) The signal u(n) is obtained after combination; the output signal of the encoding end is obtained after speech encoding the signal u(n); the signal uq(n) obtained after decoding the output signal of the encoding end is combined with the signal s ₁ (n) The signal q ₁ (n) is obtained, and the signal q ₁ (n) is processed by the first noise shaping filter to obtain the feedback signal f ₁ q ₁ (n); the signal uq(n) is combined with the speech signal s(n) to obtain the signal q ₂ (n), the feedback signal f ₂ q ₂ (n) is obtained after the second noise shaping and filtering process is performed on the signal q ₂ (n) according to the voice signal s(n).

In this process, the feedback signal f ₂ q ₂ (n) obtained after performing the second noise shaping filtering process on the signal q ₂ (n) according to the speech signal s(n) is combined with the speech signal s(n) to obtain the signal s ₁ (n), realizes the adaptive noise shaping processing of the speech signal. Further combine the signal q ₁ (n) with the first noise shaping filter to obtain the feedback signal f ₁ q ₁ (n) and s ₁ (n) to obtain the signal u(n) to realize the fixed noise shaping of the speech signal deal with.

The speech signal is first subjected to fixed noise shaping processing, and then the process of performing adaptive noise shaping processing according to the speech signal s(n) may include the following processes:

The speech signal s(n) is combined with the first feedback signal f ₁ q ₁ ′(n) to obtain the signal s ₁ ′(n), and the signal s ₁ ′(n) is combined with the second feedback signal f ₂ q ₂ ′ (n) After the combination, the signal u'(n) is obtained; the output signal of the encoding end is obtained after the speech encoding of the signal u'(n); the signal uq'(n) obtained after decoding the output signal of the encoding end is combined with the signal After combining s ₁ ′(n), the signal q ₂ ′(n) is obtained. According to the voice signal s ₁ ′(n), the signal q ₂ ′(n) is processed by the second noise shaping filter to obtain the feedback signal f ₂ q ₂ ′ (n); signal q ₁ ′(n) is obtained after combining signal uq′(n) and speech signal s(n), and feedback signal f ₁ q is obtained after signal q ₁ ′(n) is subjected to second noise shaping filter processing ₁ '(n).

In this process, the signal q ₁ ′(n) is processed by the first noise shaping filter to obtain the feedback signal f ₁ q ₁ ′(n) and the speech signal s(n) to obtain the signal s ₁ ′(n) to achieve Fixed noise shaping processing for speech signals. Further, according to the voice signal s ₁ ′(n), the signal q ₂ ′(n) is subjected to the second noise shaping and filtering process, and the feedback signal f ₂ q ₂ ′(n) is combined with the signal s ₁ ′(n) to realize Adaptive noise shaping processing for speech signals.

In the above process, the fixed noise shaping process is a rough noise shaping process on the signal spectrum envelope, and the adaptive noise shaping process based on the speech signal is a noise shaping process on the signal spectrum fine structure.

FIG. 2 is a structural diagram of the first type of noise shaping device provided by the embodiment of the present invention. As shown in FIG. 2 , the device includes: an adaptive noise shaping unit 210 , a fixed noise shaping unit 220 and a speech encoding unit 230 .

The adaptive noise shaping unit 210 is configured to perform adaptive noise shaping processing on the speech signal s(n) to obtain a signal s ₁ (n).

The fixed noise shaping unit 220 is configured to perform fixed noise shaping processing on the signal s ₁ (n) obtained after the adaptive noise shaping unit 210 performs the adaptive noise shaping processing to obtain the signal u(n).

The speech coding unit 230 is configured to perform speech coding on the signal u(n) obtained after processing by the fixed noise shaping unit 220 to obtain an output signal of the coding end.

Wherein, the adaptive noise shaping unit 210 may include: a first combination unit 211 , a second speech decoding unit 212 , a fourth combination unit 213 , a linear prediction analysis unit 214 and a second noise shaping and filtering unit 215 .

The first combining unit 211 is configured to combine the speech signal s(n) and the feedback signal f ₂ q ₂ (n) obtained by the second noise shaping and filtering unit 215 to obtain a signal s ₁ (n).

The second speech decoding unit 212 is configured to decode the output signal encoded by the speech encoding unit 230 to obtain a signal uq(n).

The fourth combination unit 213 is configured to combine the speech signal s(n) with the signal uq(n) obtained by the second speech decoding unit 212 to obtain a signal q ₂ (n).

A linear predictive analysis unit 214, configured to perform linear predictive analysis on the speech signal s(n).

The second noise shaping and filtering unit 215 is configured to perform noise shaping and filtering processing on the signal q ₂ (n) obtained by the fourth combination unit 213 according to the linear prediction analysis result obtained by the linear prediction analysis unit 214, to obtain the feedback signal f ₂ q ₂ (n).

The adaptive noise shaping unit 210 may also include: a pre-emphasis unit 216 for performing pre-emphasis processing on the speech signal s(n);

The second noise shaping and filtering unit 215 may also perform the operation of performing noise shaping and filtering on the signal q ₂ (n) obtained by the fourth combination unit 213 according to the processing result of the pre-emphasis unit 216 .

The linear prediction analysis unit 214 and the second noise shaping and filtering unit 215 may be set in the same unit.

The fixed noise shaping unit 220 may include: a second combining unit 221 , a first speech decoding unit 222 , a third combining unit 223 and a first noise shaping and filtering unit 224 .

The second combining unit 221 is configured to combine the signal s ₁ (n) obtained by the adaptive noise shaping unit 210 and the feedback signal f ₁ q ₁ (n) obtained by the first noise shaping unit to obtain a signal u(n).

The first speech decoding unit 222 is configured to decode the output signal encoded by the speech encoding unit 230 to obtain a signal uq(n).

The third combining unit 223 is configured to combine the signal s ₁ (n) obtained by the adaptive noise shaping unit 210 and the signal uq(n) obtained by the first speech decoding unit 222 to obtain a signal q ₁ (n).

The first noise shaping and filtering unit 224 is configured to perform noise shaping and filtering processing on the signal q ₁ (n) obtained by the third combining unit 223 to obtain a feedback signal f ₁ q ₁ (n).

In addition, the first speech decoding unit 222 and the second speech decoding unit 212 may also be set as one unit.

FIG. 3 is a second structure diagram of a noise shaping device provided by an embodiment of the present invention. As shown in FIG. 3 , the device includes: a fixed noise shaping unit 310 , an adaptive noise shaping unit 320 and a speech encoding unit 330 .

The fixed noise shaping unit 310 is configured to perform fixed noise shaping processing on the speech signal s(n) to obtain a signal s ₁ '(n).

The adaptive noise shaping unit 320 is configured to perform adaptive noise shaping processing on the signal s ₁ ′(n) obtained by the fixed noise shaping unit 310 to obtain a signal u′(n).

The speech encoding unit 330 is configured to perform speech encoding on the signal u'(n) obtained after processing by the adaptive noise shaping unit 320, to obtain an output signal at the encoding end.

Wherein, the fixed noise shaping unit 310 may include: a first combining unit 311 , a first speech decoding unit 312 , a fourth combining unit 313 and a first noise shaping and filtering unit 314 .

The first combining unit 311 is configured to combine the speech signal s(n) and the feedback signal f ₁ q ₁ ′(n) obtained by the first noise shaping and filtering unit to obtain a signal s ₁ ′(n).

The first speech decoding unit 312 is configured to decode the output signal encoded by the speech encoding unit 330 to obtain a signal uq'(n).

The fourth combination unit 313 is configured to combine the speech signal s(n) with the signal uq'(n) obtained by the first speech decoding unit 312 to obtain a signal q ₁ '(n).

The first noise shaping and filtering unit 314 is configured to perform noise shaping and filtering processing on the signal q ₁ ′(n) obtained by the fourth combining unit 313 to obtain a feedback signal f ₁ q ₁ ′(n).

The adaptive noise shaping unit 320 may include: a second combination unit 321 , a second speech decoding unit 322 , a third combination unit 323 , a linear prediction analysis unit 324 and a second noise shaping and filtering unit 325 .

The second combination unit 321 is configured to combine the signal s ₁ '(n) obtained by the fixed noise shaping unit 310 and the feedback signal f ₂ q ₂ (n) obtained by the second noise shaping and filtering unit 325 to obtain a signal u ₁ ( n).

The second speech decoding unit 322 is configured to decode the output signal encoded by the speech encoding unit 230 to obtain a signal uq'(n).

The third combining unit 323 is configured to combine the signal s ₁ '(n) obtained by the fixed noise shaping unit 310 and the signal uq'(n) obtained by the second speech decoding unit 322 to obtain a signal q ₂ '(n).

A linear predictive analysis unit 324, configured to perform linear predictive analysis on the signal s ₁ ′(n) obtained by the fixed noise shaping unit 310.

The second noise shaping and filtering unit 325 is used to perform noise shaping and filtering processing on the signal q ₂ '(n) obtained by the third combination unit 323 according to the linear prediction analysis result obtained by the linear prediction analysis unit 324, to obtain the feedback signal f ₂ q ₂ '(n).

The adaptive noise shaping unit 320 may also include: a pre-emphasis unit 326, configured to perform pre-emphasis processing on the signal s ₁ '(n);

The second noise shaping and filtering unit 325 may also perform the operation of performing noise shaping and filtering on the signal q ₂ ′(n) obtained by the third combination unit 323 according to the processing result of the pre-emphasis unit 216 .

The linear prediction analysis unit 324 and the pre-emphasis unit 326 may be implemented in the same unit.

The first speech decoding unit 312 and the second speech decoding unit 322 may also be configured as one unit.

The structure diagram of the noise shaping device shown in Fig. 2 is corresponding to performing adaptive noise shaping processing according to the speech signal first, and then performing fixed noise shaping processing, and the composition structure diagram of the noise shaping device shown in Fig. 3 is the same as first performing The fixed noise shaping process is corresponding to the adaptive noise shaping process.

In order to make the present invention more clear, the noise shaping device with the structure shown in FIG. 2 and its corresponding method will be described in detail below with a specific embodiment. Fig. 4 is a schematic structural diagram of a specific noise shaping device provided by an embodiment of the present invention. In Fig. 4, the first speech decoding unit and the second speech decoding unit are set as a local speech decoder, and the pre-emphasis unit and linear prediction analysis unit are set as a pre-emphasis and linear prediction analyzer. The following is the feedback process for implementing noise shaping using this specific device:

1) Both the speech signal s(n) and the second feedback signal f ₂ q ₂ (n) are input into the first adder, and combined to output the signal s ₁ (n).

The signal combination performed by the first adder may be the addition or subtraction of the speech signal s(n) and the second feedback signal f ₂ q ₂ (n). When adding or subtracting, it will affect the selection of the noise shaping filter mode of the second noise shaping filter.

2) Both the signal s ₁ (n) output by the first adder and the first feedback signal f ₁ q ₁ (n) are input into the second adder, and combined to output a signal u(n).

The signal combination performed by the second adder may also be the addition or subtraction of the speech signal s ₁ (n) and the first feedback signal f ₁ q ₁ (n). When adding or subtracting, it will affect the selection of the noise shaping filter mode of the first noise shaping filter.

3) After inputting the signal u(n) output by the second adder into the speech coder for speech coding, the output signal of the coding end is obtained. The output signal is the signal that needs to be transmitted and finally reaches the decoding end. The speech signal can be restored by decoding the signal at the decoding end. The restored speech signal is the speech signal after noise shaping, which improves the subjective auditory quality of the speech signal.

4) At the encoding end, the signal output by the speech encoder is decoded by the local speech decoder to output the signal uq(n).

5) Input the signal uq(n) output by the local speech decoder and the signal s ₁ (n) output by the first signal adder into the third adder, and output signal q ₁ (n) after combining.

The combination of signal uq(n) and signal s ₁ (n) by the third adder is to subtract the two signals, and the selection of the sign of the absolute value after subtraction will also affect the noise shaping filtering of the first noise shaping filter way of selection.

6) Input the signal q ₁ (n) output by the third adder into the first noise shaping filter for processing, and output the feedback signal f ₁ q ₁ (n).

The processing method of the first noise shaping filter is preset, that is, it is independent of the speech signal. Once the setting is completed, the noise shaping method _F1 is fixed for different speech signals, and its main function is to shape the spectrum envelope of the speech signal, that is, to perform rough noise shaping on the flat spectrum of the quantization noise , to suppress high-frequency noise.

When the combination mode of the second adder is addition, and the combination mode of the third adder is s ₁ (n) minus uq(n), the quantization noise produced by the speech encoder and the local speech decoder is: △( n)=u(n)-uq(n), the quantization noise after the first noise shaping filter is q ₁ (n)=s ₁ (n)-uq(n). After performing z-transformation on the above two formulas, the relationship between the quantization noise after the first noise shaping filter and the quantization noise generated by the speech encoder and the local speech decoder in the frequency spectrum can be obtained as follows: $Q_1\left(z\right)=\frac{1}{1+f_1\left(z\right)}\mathrm{\Δ}\left(z\right)---\left(1\right)$

When presetting the processing mode of the first noise shaping filter, formula (1) can be used, and the expression _F1 of the processing mode can be selected according to the performance of the signal to be obtained. For example, if you want to reduce the high-frequency noise energy, you can choose according to formula (1), and you can set _F 1 =-0.8z ^-1 , and you can get good results. Of course, the selected F ₁ is not only a unique expression, but can use formula (1) to select according to actual needs, that is to say, the first noise shaping filter can have multiple designs, zero-pole, all-zero , all-pole filters that can obtain the desired signal performance can be selected.

When the combination of the second adder and/or the third adder changes, the same method can be used to obtain different quantization noise after the first noise shaping filter and the speech encoder and the local speech decoder. The relationship of quantization noise on the frequency spectrum, that is, the second adder and/or the third adder adopts different combinations, and the sign in (1) changes slightly, for example, when the combination of the second adder is subtraction, And when the combination of the third adder is uq(n) minus s ₁ (n), the formula (1) becomes: $Q_1\left(z\right)=\frac{1}{1+f_1\left(z\right)}\mathrm{\Δ}\left(z\right),$ Other situations will not be repeated here.

7) Input the signal uq(n) output by the local speech decoder and the speech signal s(n) into the fourth adder, combine them and output the signal q ₂ (n).

The combination method of the fourth adder to the signal uq(n) and the voice signal s(n) is to subtract the two signals, and the selection of the absolute value sign of the data obtained after the subtraction will also affect the noise shaping of the second noise shaping filter Selection of filtering method.

8) The voice signal is input into the pre-emphasis and linear predictive analyzer for pre-emphasis and linear predictive analysis processing, and the second noise shaping filter utilizes the processing results output by the pre-emphasis and linear predictive analysis processor to determine its own adaptive noise shaping filtering method.

The adaptive noise shaping and filtering processing method is dynamically changed along with the frequency spectrum of the speech signal.

9) Input the signal q ₂ (n) output by the fourth adder into the second noise shaping filter, and the second noise shaping filter uses its own adaptive noise shaping filter processing method to process the signal q ₂ (n), and output Feedback signal f ₂ q ₂ (n).

When the combination mode of the first adder is addition, and the combination mode of the fourth adder is to subtract the signal uq(n) from the speech signal s(n), the quantization noise produced by the speech encoder and the local speech decoder is still assumed is: △(n)=u(n)-uq(n), the quantization noise after the second noise shaping filter is q ₂ (n)=s(n)-uq(n), and because q ₂ ( n)=q ₁ (n)-f ₂ q ₂ (n-1), after carrying out z transformation to above-mentioned two formulas, can draw the quantization noise after the second noise shaping filter and speech coder and local The quantization noise produced by the speech decoder is related on the frequency spectrum as:

${\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

The selection of the shaping filtering mode of the second noise shaping filter is adaptively changed according to the change of the speech signal, and the expression _F2 of its shaping filtering mode is selected according to the results of pre-emphasis and linear prediction analysis. Its main function is to perform noise shaping on the fine spectrum structure of the speech signal, that is, to perform finer noise shaping on the spectrum between harmonics without changing the general shape of the previous noise shaping spectrum, thereby further improving subjective hearing quality.

Because the noise power spectrum decreases with the increase of frequency, most of its energy is concentrated in the low frequency range. In order to suppress the low frequency noise of the speech signal, a pre-emphasis method is adopted. Pre-emphasis processing is to use the difference between the characteristics of the speech signal and the noise to effectively process the signal, and suppress the low-frequency noise by emphasizing the high-frequency component of the speech signal. The specific pre-emphasis method is existing in the prior art. method, which will not be repeated here.

In order to make the processing mode of the second noise shaping filter closer to the spectrogram of the speech signal, the embodiment of the present invention adopts the method of linear predictive analysis so that each sampling value of the speech signal can be approximated by a linear combination of several sampling values in the past, Specifically, a unique set of weighting coefficients used in the linear combination is determined as the linear prediction result by minimizing the mean square error between the actual speech signal sample value and the linear prediction sample value. The noise between the middle and high frequency signal harmonics in the speech signal can be suppressed by the method of linear predictive analysis. This is also an existing method in the prior art, and will not be repeated here.

In addition, the linear predictive analysis may be performed on the pre-emphasized speech signal, and the expression F2 of the second noise shaping filtering processing method is determined according to the prediction result obtained by the linear predictive analysis. It is also possible to only perform linear predictive analysis on the speech signal, and determine the expression F ₂ of the processing mode of the second noise shaping filter according to the prediction result of the linear predictive analysis. After linear predictive analysis, m weighting coefficients α _i , i=1, . . . , m are obtained. Then, the expression F ₂ is determined based on the formula (2) and the performance of the desired signal. For example, you can make $f_2=\frac{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{a}_i(1-{\mathrm{\&alpha;}}^i)z^{-i}}{1+\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{a}_i{\mathrm{\&alpha;}}^i{z}^{-i}}.$ Similarly, the selected F ₂ does not have a unique expression, and can use formula (2) to select according to actual needs, that is to say, the second noise shaping filter is not limited to a specific return, zero-pole, full All zero-point and all-pole filters that can meet the performance requirements can be selected.

Fig. 5 is a schematic structural diagram of another specific noise shaping device provided by an embodiment of the present invention. The specific structure in Fig. 5 is a specific example of the noise shaping device and the corresponding method shown in Fig. 3. The difference between the implementation of this structure and Fig. 4 is that the adaptive noise shaping process is performed first, and then the fixed noise Plastic processing, the specific processing process will not be repeated.

It can be seen from the above description that in the method and device provided by the embodiments of the present invention, the speech signal is first subjected to adaptive noise shaping processing, and then fixed noise shaping processing is performed, and the speech signal after the fixed noise shaping processing is performed. Encoding to form an output signal at the encoding end; or, perform fixed noise shaping processing on the speech signal first, and then perform adaptive noise shaping processing, and perform speech encoding on the signal after adaptive noise shaping processing to form an output signal at the encoding end. Since both the fixed noise shaping process and the adaptive noise shaping process are noise shaping processes on the speech signal itself, and do not involve the speech signal and the residual signal of the prediction filter, there is no need to make any changes at the decoding end The noise-shaped speech signal can be decoded and restored.

Moreover, the embodiment of the present invention provides a specific two-layer structure shaping filter device and method that combines fixed filter shaping and adaptive filter shaping. Perform rough noise shaping on the flat spectrum, determine the way of adaptive filtering and shaping by pre-emphasizing and linear predictive analysis on the speech signal, and perform adaptive filtering and shaping in real time according to changes in the speech signal, so as not to change the general shape of the speech signal spectrum Under the condition of , the fine structure of the speech signal spectrum is shaped, so as to further improve the subjective auditory quality. At the same time, pre-emphasis and linear predictive analysis are used to determine the adaptive filter shaping method, which can well suppress the low-frequency noise of the speech signal, and at the same time suppress the middle and high-frequency noise of the speech signal, so that the auditory quality of the speech signal is further improved. improved.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the present invention. within the scope of protection.

Claims (20)

1. A method of noise shaping, the method comprising:

performing adaptive noise shaping processing on the speech signal s (n) to obtain a signal s₁(n)；

For the signal s₁(n) performing fixed noise shaping processing to obtain a signal u (n);

and carrying out voice coding on the signal u (n) to form an output signal of a coding end.

2. According to the claimsThe method of claim 1, wherein the adaptive noise shaping process is performed on the speech signal s (n) to obtain the signal s₁(n) comprises: performing voice decoding on the signal output by the encoding end to obtain a signal uq (n), and combining the signal uq (n) with the voice signal s (n) to obtain a signal q₂(n) applying the signal q to the speech signal s (n)₂(n) performing a second noise shaping filtering process to obtain a feedback signal f₂q₂(n) combining the speech signal s (n) with a second feedback signal f₂q₂(n) combining to obtain a signal s₁(n)。

3. Method according to claim 1, characterized in that said signal s is measured₁(n) performing a fixed noise shaping process to obtain a signal u (n) comprising: performing speech decoding on the signal output by the encoding end to obtain a signal uq (n), and combining the signal uq (n) with the signal s₁(n) are combined to obtain a signal q₁(n) converting said signal q₁(n) obtaining a feedback signal f after a first noise shaping filtering process₁q₁(n) converting said signal s₁(n) and the first feedback signal f₁q₁(n) combined to give the signal u (n).

4. Method according to claim 2, characterized in that said signal q is derived from a speech signal s (n)₂(n) performing the second noise-shaping filtering process includes: performing linear predictive analysis on the speech signal s (n), determining a noise shaping filter processing mode of a second noise shaping filter processing in real time according to a result of the linear predictive analysis processing, and performing linear predictive analysis on the signal q by using the determined processing mode₂(n) performing a second noise shaping filtering process; or,

pre-emphasis and linear predictive analysis processing are performed on the speech signal s (n), a noise shaping filter processing mode of a second noise shaping filter processing is determined in real time according to a result of the pre-emphasis and linear predictive analysis processing, and the signal q is processed by using the determined processing mode₂(n) performing a second noise shaping filtering process.

5. The method according to claim 2, wherein said combining the signal uq (n) with the speech signal s (n) comprises: subtracting the speech signal s (n) from the signal uq (n) or subtracting the signal uq (n) from the speech signal s (n);

mixing the speech signal s (n) with a second feedback signal f₂q₂(n) combining comprises: mixing the speech signal s (n) with a second feedback signal f₂q₂(n) adding or subtracting.

6. Method according to claim 3, characterized in that the signal uq (n) is compared with the signal s₁(n) combining comprises: subtracting the signal s from the signal uq (n)₁(n) or the signal s₁(n) subtracting the signal uq (n);

the signal s is measured₁(n) and the first feedback signal f₁q₁(n) combining comprises: the signal s is measured₁(n) and the first feedback signal f₁q₁(n) adding or subtracting.

7. A method of noise shaping, the method comprising:

performing fixed noise shaping processing on the speech signal s (n) to obtain a signal s₁′(n)；

For the signal s₁'(n) performing adaptive noise shaping processing to obtain a signal u' (n);

and carrying out voice coding on the signal u' (n) to form an output signal of a coding end.

8. The method of claim 7, wherein the speech signal s (n) is subjected to a fixed noise shaping process to obtain a signal s₁' (n) includes: performing voice decoding on the signal output by the encoding end to obtain a signal uq' (n), and performing voice decoding on the signal uq' (n) is combined with the signal s (n) to obtain a signal q₁' (n) converting said signal q into a signal q₁' (n) obtaining a feedback signal f after a first noise shaping filtering process₁q₁' (n) comparing said signal s (n) with a first feedback signal f₁q₁' (n) are combined to obtain a signal s₁′(n)。

9. Method according to claim 7, characterized in that said signal s is measured₁'(n) obtaining the signal u' (n) by performing adaptive noise shaping processing includes: performing speech decoding on the signal output by the encoding terminal to obtain a signal uq '(n), and combining the signal uq' (n) with the signal s₁' (n) are combined to obtain a signal q₂' (n) based on said signal s₁' (n) pairs of signals q₂' (n) performing second noise shaping filtering to obtain feedback signal f₂q₂' (n) converting said signal s₁' (n) and a second feedback signal f₂q₂'(n) are combined to obtain a signal u' (n).

10. Method according to claim 9, characterized in that, depending on the signal s₁' (n) pairs of signals q₂' (n) the performing of the second noise shaping filtering process includes: for the signal s₁' (n) performing a linear prediction analysis process, determining a noise shaping filter process mode of a second noise shaping filter process in real time from a result of the linear prediction analysis process, and performing the linear prediction analysis process on the signal q using the determined noise shaping filter process mode₂' (n) performing a second noise shaping filtering process; or,

for the signal s₁' (n) performing pre-emphasis and linear prediction analysis processing, determining a noise shaping filter processing mode of a second noise shaping filter processing in real time based on a result of the pre-emphasis and linear prediction analysis processing, and performing the pre-emphasis and linear prediction analysis processing on the signal q in accordance with the determined noise shaping filter processing mode₂' (n) second noise shaping filtering processing is performed.

11. The method of claim 8, wherein combining the signal uq' (n) with the signal s (n) comprises: subtracting the signal s (n) from the signal uq '(n), or subtracting the signal uq' (n) from the signal s (n);

the signal s (n) is compared with a first feedback signal f₁q₁' (n) the combination comprises: mixing s (n) with a first feedback signal f₁q₁' (n) are added or subtracted.

12. Method according to claim 9, characterized in that the signal uq' (n) is related to the signal s₁The combination of' (n) includes: subtracting said signal s from said signal uq' (n)₁' (n) or, the signal s₁'(n) subtracting said signal uq' (n);

the signal s is measured₁' (n) and a second feedback signal f₂q₂' (n) the combination comprises: the signal s is measured₁' (n) and a second feedback signal f₂q₂' (n) are added or subtracted.

13. An apparatus for noise shaping, the apparatus comprising:

an adaptive noise shaping unit for performing adaptive noise shaping processing on the speech signal s (n) to obtain a signal s₁(n)；

A fixed noise shaping unit for performing adaptive noise shaping processing on the signal s obtained by the adaptive noise shaping unit₁(n) performing fixed noise shaping processing to obtain a signal u (n);

and the voice coding unit is used for carrying out voice coding on the signal u (n) obtained after the fixed noise shaping unit processes to obtain an output signal of a coding output end.

14. The apparatus of claim 13, wherein the adaptive noise shaping unit comprises:

the second voice decoding unit is used for decoding the output signal obtained after the voice coding unit codes to obtain a signal uq (n);

a fourth combining unit, configured to combine the speech signal s (n) with the signal uq (n) obtained by the second speech decoding unit to obtain a signal q₂(n)；

A linear prediction analysis unit for performing linear prediction analysis on the speech signal s (n);

a second noise shaping filter unit for processing the signal q obtained by the fourth combination unit according to the linear prediction analysis result obtained by the linear prediction analysis unit₂(n) performing noise shaping filtering processing to obtain a feedback signal f₂q₂(n)；

A first combination unit for combining the speech signal s (n) with the feedback signal f obtained by the second noise shaping filter unit₂q₂(n) combining to obtain a signal s₁(n)。

15. The apparatus of claim 14, wherein the adaptive noise shaping unit further comprises: a pre-emphasis unit, configured to perform pre-emphasis processing on the speech signal s (n);

the second noise shaping and filtering unit may further perform the operation of performing noise shaping and filtering processing on the signal obtained by the fourth combining unit according to the processing result of the pre-emphasis unit.

16. The apparatus of claim 13, wherein the fixed noise shaping unit comprises:

the first voice decoding unit is used for decoding the output signal obtained after the voice coding unit codes to obtain a signal uq (n);

a third combination unit for combining the signal s obtained by the adaptive noise shaping unit₁(n) is combined with the signal uq (n) obtained by the first speech decoding unit to obtain a signal q₁(n)；

A first noise shaping filter unit forFor the signal q obtained for the third combination unit₁(n) performing noise shaping filtering processing to obtain a feedback signal f₁q₁(n)；

A second combination unit for combining the signal s obtained by the adaptive noise shaping unit₁(n) and the feedback signal f obtained by the first noise shaping unit₁q₁(n) are combined to give the signal u (n).

17. An apparatus for noise shaping, the apparatus comprising:

a fixed noise shaping unit for performing fixed noise shaping processing on the speech signal s (n) to obtain a signal s₁′(n)；

An adaptive noise shaping unit for shaping the signal s obtained by the fixed noise shaping unit₁'(n) performing adaptive noise shaping processing to obtain a signal u' (n);

and a voice coding unit for performing voice coding on the signal u' (n) obtained by the adaptive noise shaping unit to obtain an output signal of a coding end.

18. The apparatus of claim 17, wherein the fixed noise shaping unit comprises:

a first speech decoding unit, configured to decode an output signal encoded by the speech encoding unit to obtain a signal uq' (n);

a fourth combining unit, configured to combine the speech signal s (n) with the signal uq' (n) obtained by the first speech decoding unit to obtain a signal q₁′(n)；

A first noise shaping and filtering unit for shaping the signal q obtained by the fourth combination unit₁' (n) performing noise shaping filtering processing to obtain feedback signal f₁q₁′(n)；

A first combination unit for combining the speech signal s (n) with the feedback signal f obtained by the first noise shaping filter unit₁q₁' (n) are combined to obtain a signal s₁′(n)。

19. The apparatus of claim 17, wherein the adaptive noise shaping unit comprises:

a second speech decoding unit, configured to decode the output signal obtained by encoding by the speech encoding unit, so as to obtain a signal uq' (n);

a third combination unit for combining the signal s obtained by the fixed noise shaping unit₁'(n) is combined with the signal uq' (n) obtained by the second speech decoding section to obtain a signal q₂′(n)；

A linear prediction analysis unit for performing linear prediction on the signal s obtained by the fixed noise shaping unit₁' (n) performing a linear predictive analysis;

a second noise shaping filter unit for processing the signal q obtained by the third combination unit according to the linear prediction analysis result obtained by the linear prediction analysis unit₂' (n) performing noise shaping filtering processing to obtain feedback signal f₂q₂′(n)；

A second combination unit for combining the signal s obtained by the fixed noise shaping unit₁' (n) and the feedback signal f obtained by the second noise shaping filter unit₂q₂'(n) are combined to obtain a signal u' (n).

20. The apparatus of claim 19, wherein the adaptive noise shaping unit further comprises: a pre-emphasis unit for pre-emphasizing the signal s obtained by the fixed noise shaping unit₁' (n) performing pre-emphasis processing;

the second noise shaping and filtering unit is further configured to execute the signal q obtained by the third combining unit according to the processing result of the pre-emphasis unit₂' (n) an operation of performing a noise shaping filtering process.

Images