JP2000099096A - Component separation method of voice signal, and voice encoding method using this method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for separating surely a component of an input voice signal to a first component in which a voice is the subject and a second component in which background noise is the subject for each prescribed time unit. SOLUTION: A component separating section 100 applying this component separating method has noise suppression processing sections 101, 102 performing a first noise suppression processing of which suppression quantity is relatively small for an input voice signal and a second noise suppression processing of which suppression quantity is relatively large and outputting voice component extracting signals S1, S2, a subtracter 103 subtracting the signal S2 from an input voice signal and obtaining a noise component extracting signal S3, a state deciding section 104 deciding which state the input voice signal is, a first state in which a voice component is main, a second state in which a background noise component is main, or a third state in which component other than the above is main, from the signals S2, S3 for each prescribed time, and a selecting section 105 selecting any of signals S1, S2, S3, and all zero signal S4 conforming to a decided result of the state deciding section 104 as first and second components.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio signal component for separating an input audio signal into a first component mainly composed of speech and a second component mainly composed of background noise every predetermined time unit. The present invention relates to a separation method and a speech encoding method for compressing and encoding a speech signal including background noise with high efficiency in a form as close to the original sound as possible.

[0002]

2. Description of the Related Art Conventional low bit rate speech coding is performed by a coding method which incorporates a model of a speech generation process for the purpose of efficiently coding a speech signal. Among such audio coding methods, particularly in recent years, C
Speech coding methods based on the ELP method have been widely used. With this CELP based speech coding method,
A speech signal input in an environment where there is almost no background noise matches the coding model, so that the coding efficiency is high and the coding can be performed with a relatively small deterioration in sound quality. .

However, when a CELP-based speech coding method is used for a speech signal input under the condition that the level of background noise is large, the reproduced output signal has a very unstable feeling of background noise and is very unstable. It is known that it will sound unpleasant. Such a tendency is particularly remarkable when the encoding bit rate is 8 kbps or less.

In order to reduce such a problem, CELP coding is performed using a source signal having higher noise in a time section determined to be background noise, so that sound quality degradation in the background noise section is reduced. Methods to improve it have also been proposed. Although the sound quality of the background noise section is somewhat improved by using such a method, the sound generation process model of synthesizing the sound by passing the sound source signal through the synthesis filter is still used, so that the background of the original sound is still used. There is no change in the tendency of noise to be different from noise, and there is a problem that the effect of improvement is small.

[0005]

As described above, in the conventional speech encoding method, when an input speech signal is encoded under the condition that the background noise level is large, the reproduced output signal feels like background noise. There was a problem that the sound was greatly changed and the sound was very unstable and unpleasant.

In order to solve this problem, an input speech signal is separated into a component mainly composed of speech and a component mainly composed of background noise, and based on different models suited to the properties of the respective components of speech and background noise. It is conceivable to use a different encoding method. In this case, it is necessary to correctly separate the input audio signal into a component mainly composed of voice and a component mainly composed of background noise, but no effective method for such component separation has been found in the past.

[0007] The present invention has been made in view of the above points, and it is possible to reliably convert an input audio signal into a first component mainly composed of voice and a second component mainly composed of background noise every predetermined time unit. Speech signal suitable for realizing low-rate speech coding that can reproduce speech as close as possible to the original sound, including background noise, using such a component separation method. It is to provide an encoding method.

[0008]

In order to solve the above-mentioned problems, a method for separating components of an audio signal according to the present invention comprises the steps of dividing an input audio signal into a first component mainly composed of a speech and a background noise every predetermined time unit. When the component is separated into the main component and the second component, a signal obtained by performing a first noise suppression process with a relatively small amount of suppression on the input audio signal is set as a first component, and On the other hand, there is provided a separation mode in which a signal obtained by performing a second noise suppression process having a relatively large amount of suppression and which is obtained by subtracting a signal obtained from the input audio signal is used as a second component. This separation mode is effective when the input voice signal contains both voice components and noise components.

When the first noise suppression processing with a relatively small amount of suppression is performed on an input speech signal, a speech component with a small loss of information is obtained. Although the decoded speech component obtained by encoding / decoding this speech component is perceived to be degraded at the portion of the mixed noise component, since there is no loss of information of the speech component, high quality can be obtained by paying attention to the speech component. A decoded speech component can be obtained. On the other hand, the noise component obtained by performing the second noise suppression process with a relatively large amount of suppression on the input speech signal accurately represents the noise component although the speech component is mixed. By performing the encoding / decoding process, a high-quality decoding noise component can be obtained.
Therefore, when both the speech component and the noise component separated in this way are combined, the sense of incompatibility of the mixed noise component, which is a problem in the decoded speech component generated in the speech encoding / decoding process, is reduced. It is masked by the decoding noise component generated in the encoding / decoding process, and becomes inaudible, so that high-quality decoded speech can be finally obtained.

According to a more specific audio signal component separation method according to the present invention, a first signal obtained by performing a first noise suppression with a relatively small amount of suppression on an input audio signal;
A second signal obtained by performing a second noise suppression process with a relatively large amount of suppression on the input audio signal and a third signal obtained by subtracting the second signal from the input audio signal are generated. . The first audio signal mainly includes an audio component.
State, a second state mainly including a background noise component, and a third state other than the first and second states including both a voice component and a background noise component for each predetermined time unit. When it is determined to be the first state, the input audio signal is output as it is as the first component, and a predetermined signal is output as the second component, and the second state is determined. When it is determined, a predetermined signal is output as a first component, and an input audio signal is output as a second component. When it is determined to be in the third state, the first component is output. And a third signal is output as the second component. Here, the third state is the separation mode described above. The predetermined signal is, for example, an all-zero signal.

When it is determined that the input audio signal is in the first state, the input audio signal is output as it is as a first component, and a predetermined signal is output as a second component. If the second state is determined, the first state
The second signal is output as the second component, the third signal is output as the second component, and when the third state is determined, the first signal is output as the first component. , A third signal may be output as the second component.

Further, when the input audio signal is in the first state,
It is desirable that the process of determining which of the state (3) and the state (3) includes a process of checking the magnitude of the pitch periodicity of the noise component extraction signal.

According to such an audio signal component separation method, when the input audio signal is in the first state mainly including the audio component, the input audio signal is not subjected to the noise suppression processing and is not subjected to the noise component main processing. 1 is output as a component, and is encoded by an encoding method suitable for the audio component, so that the quality degradation of the decoded audio signal due to unnecessary noise suppression processing on the audio component is avoided.

Further, the magnitude of the pitch periodicity of the noise component extraction signal is examined. If this value is less than a certain threshold value, it is determined that the input speech signal is in the second state mainly including the background noise component, and the background noise component is determined. Output as the second component of the subject,
This is encoded by an encoding method suitable for the background noise component. On the other hand, when the magnitude of the pitch periodicity is equal to or larger than the threshold,
Assuming that the input audio signal is in the first state mainly including the audio component, the input audio signal is output as the first component, and this is encoded by an encoding method suitable for the audio component. When the noise component extraction signal has a large pitch periodicity as described above, the noise component extraction signal is encoded by an encoding method suitable for the audio component, so that the noise component extraction signal mainly includes the background noise component. The generation of unpleasant noise due to encoding by an encoding method suitable for the component is avoided.

That is, a second signal mainly composed of a background noise component
In general, an encoder based on an encoding method suitable for the component cannot efficiently represent the periodicity, and may generate unpleasant noise when a periodic signal is input. In such a case, the generation of unpleasant noise is achieved by encoding the input audio signal using an encoder capable of correctly expressing the periodicity based on an encoding method suitable for the first component mainly including the audio component. Can be prevented.

In the speech encoding method according to the present invention, the first component mainly composed of speech and the second component mainly composed of background noise are converted into the input speech signal by the above-described component separation method according to the present invention every predetermined time unit. A plurality of bits for which the bit assignment of each component is predetermined based on the first and second components or on the basis of the determination result of the state determined at the time of the component separation. A selection is made from among allocation candidates, and the first and second components are respectively coded by different predetermined coding methods under this bit allocation.
And coded data of the second component and information of bit allocation are output as transmission coded data.

In the CELP coding, as described above, when an input speech signal is encoded under the condition that the level of the background noise is large, the feeling of the background noise of the reproduced speech signal changes greatly, which is very unstable and unpleasant. It sounds like a feeling. This is because the background noise has a completely different model from the speech signal that CELP is good at, and it is desirable to encode the background noise by a method corresponding to it.

In the speech encoding method according to the present invention, the input speech signal is separated into a first component mainly composed of speech and a second component mainly composed of background noise for each predetermined time unit, so that speech and speech are separated. By using an encoding method based on a different model suited to the nature of each component of the background noise, the efficiency of the overall encoding is improved. At this time, based on the first component and the second component, a bit allocation that can more efficiently encode each component is selected from predetermined candidates, and the first component and the second component are determined. By encoding
The input audio signal can be encoded with high efficiency while keeping the overall bit rate low.

On the other hand, in the speech decoding method according to the present invention,
In order to decode the transmission encoded data obtained by performing the encoding as described above and reproduce the audio signal, the encoded data of the first component mainly composed of audio is input from the input transmission encoded data. And the bit allocation information of the second component coded data mainly composed of the background noise and the bit allocation information of the first and second component coded data, and decodes the bit allocation information to obtain the first and second bits. A bit assignment of the encoded data of the second component is obtained, and the encoded data of the first and second components are decoded under the bit assignment to reproduce the first and second components. The second component can be combined to produce a final output audio signal.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Before describing a specific embodiment of the present invention, a basic concept of a method for separating components of an audio signal according to the present invention will be described. In speech encoding / decoding described later, if accurate component separation between a speech component and a noise component cannot be performed at the time of component separation of a speech signal, the decoded speech will have a large perceptual deterioration. However, it is generally difficult to always perform accurate component separation in a configuration that uses noise suppression processing. The problem of the separation of the speech component and the noise component by the noise suppression processing will be described with reference to FIG.

FIG. 1A shows the relationship between the information on the audio component and the information on the noise component contained in the input audio signal. Here, the horizontal axis represents time, and the vertical axis represents the magnitude of information of each component included in the input audio signal. The noise component appears almost uniformly at all times, and the voice component is buried in the noise component at a small part (in this case, the beginning and end) of the information.

FIG. 1B and FIG. 1C show the result of component separation using noise suppression processing from the input speech signal. FIG. 1B shows a speech component after component separation, and FIG. 1C shows a noise component. Paying attention to FIG. 1B, although the noise component is hardly mixed and removed neatly, a part of the voice component is lost. On the other hand, FIG.
In FIG. 1C, although the noise components are separated, FIG.
The voice component lost in (b) is mixed. If the audio component is encoded in such a situation, the audio component which is the input of the audio encoding has already been degraded by loss of information, so that the quality of the decoded audio obtained after the encoding is reduced.

On the other hand, a method is conceivable in which the amount of suppression in the noise suppression processing is reduced to reduce the loss of information of voice components, but in that case, another problem occurs. This problem will be described with reference to FIG. Here, FIG.
As in (a), the relationship between the information on the audio component and the information on the noise component included in the input audio signal is shown. FIG. 2B shows an audio component when the input audio signal shown in FIG. 2A is subjected to component separation using noise suppression processing with a small amount of suppression, and FIG. 2C shows the audio component. Focusing on FIG. 2 (b),
Since the amount of suppression in the noise suppression processing is small, the disappearance of the voice component is small, but the noise component is large in this time. When such an audio component is encoded, the sense of the decoded noise component is greatly changed, and the sound becomes very unstable and unpleasant.

Therefore, the present invention newly employs a second noise suppression process having a smaller amount of suppression than the first noise suppression process,
Provided is a component separation method capable of avoiding quality deterioration. Less than,
The principle of the audio signal component separation method according to the present invention will be described with reference to FIG. For the input audio signal shown in FIG.
As described above, by performing the noise suppression process (first noise suppression process) with a relatively small amount of suppression,
As shown in (b), it is possible to obtain an audio component with a small loss of information. Although the decoded speech component obtained by encoding / decoding this speech component is perceived to be degraded at the portion of the mixed noise component, since there is no loss of information of the speech component, high quality can be obtained by paying attention to the speech component. A decoded speech component can be obtained.

On the other hand, a noise component obtained by performing a noise suppression process (second noise suppression process) having a relatively large suppression amount on the input speech signal shown in FIG. 3A is shown in FIG. As shown in (1), the noise component is accurately represented even though the speech component is mixed, and a high-quality decoded noise component can be obtained by subjecting this signal to noise encoding / decoding.

Therefore, when both the separated speech component and the noise component are combined as shown in FIGS. 3B and 3C, there is a problem that the decoded speech component generated in the speech encoding / decoding process causes a problem. The uncomfortable feeling of the noise component is masked by the decoded noise component generated by the noise encoding / decoding process, and becomes inaudible, so that a high-quality decoded speech can be finally obtained.

Hereinafter, a more specific embodiment of the method for separating components of an audio signal according to the present invention based on the above-described principle will be described. (First Embodiment) FIG. 4 shows a component separation unit 10 to which a method for separating components of an audio signal according to a first embodiment of the present invention is applied.
0 is shown. The component separation unit 100 separates the input audio signal into a first component mainly composed of voice and a second component mainly composed of background noise every predetermined time unit,
First noise suppression processing unit 101, second noise suppression processing unit 10
2, subtractor 103, state determination unit 104, and selection unit 10
Consists of five.

The first noise suppression processing unit 101 performs a noise suppression process with a relatively small amount of suppression on the input speech signal,
The first signal S1 is output. Second noise suppression processing unit 102
Performs a noise suppression process with a relatively large amount of suppression on an input audio signal, and outputs a second signal S2. Subtractor 10
3 subtracts the second signal S2 from the input audio signal and outputs a third signal S3.

In this example, the state judging unit 104 determines from the second signal S2 and the third signal S3 that the input audio signal is in the first state mainly composed of an audio component and in the second state mainly composed of a background noise component. , And a first containing both a speech component and a background noise component,
It is determined for each predetermined time unit whether the state is the third state other than the second state.

As shown in FIG. 5, the state determination unit 104 includes an SNR determination unit 111, a periodicity determination unit 112, an estimated energy determination unit 113, and each of the determination units 111, 112, and 1,
It is composed of an overall judgment section 114 which makes an overall judgment from the judgment results of No. 13 and outputs a state judgment result of the input audio signal. The determination result of this state determination unit 104 is
05.

In the selection section 105, when the input voice signal is determined to be in the first state by the state determination section 104, the first
And output the input audio signal as it is,
A predetermined signal S4 predetermined as a second component is output, and when the input audio signal is determined to be in the second state, the predetermined signal S4 or the second signal predetermined as the first component is output. S2 is output, and the input audio signal or the third signal S3 is output as the second component. When the input audio signal is determined to be in the third state, the first signal S1 is output as the first component. And outputs the third component as the second component.
Is output.

The state determination unit 1 in the component separation unit 100
04 is also input to a bit allocation selection unit 120 and a multiplexing unit 150 described later. Since the state determination unit 104 determines which of the first, second, and third states the input audio signal is in, the bit allocation selection unit 120 determines whether the input audio signal is a first audio signal mainly composed of audio. In the state, more bits are assigned to the speech encoder 130. When the input speech signal is in the second state mainly composed of background noise, more bits are assigned to the noise encoder 140. Is a third state that includes a large amount of both speech and background noise, bits corresponding to the information amounts of the speech component and the noise component are assigned to the speech encoding unit 130 and the background noise encoding unit 140.

In addition, the result of the determination by the state determination unit 104 indicates the information of the bit allocation in the bit allocation selection unit 120, and this is used as a bit allocation code by the multiplexing unit 150 and the audio coding unit 130 and It may be multiplexed with the encoded data from the noise encoding unit 140.

Next, the processing procedure of the component separation unit 100 will be described with reference to the flowchart shown in FIG. This process
This is performed for each frame of the input audio signal. Component separation unit 10
For the first noise suppression processing unit 101 and the second noise suppression processing unit 102 at 0, existing techniques can be used.
Here, the case where the spectral subtraction method is used as described above will be described.

First, a first signal (hereinafter, referred to as a first audio component extraction signal) S1 is subjected to noise suppression processing on an input audio signal in a first noise suppression processing unit 101 having a relatively small amount of suppression. Generated by the second noise suppression processing unit 1
02, a second signal (hereinafter, referred to as a second audio component extraction signal) S2 is generated by performing a noise suppression process with a relatively large amount of suppression on the input audio signal. A second audio component extraction signal S2 from the signal
To generate a third signal (hereinafter, referred to as a noise component extraction signal) S3 (step S1001).

Next, the state of the input audio signal from the second audio component extraction signal S2 and the noise component extraction signal S3 by the state determination unit 104, that is, the first state in which the input audio signal mainly includes the audio component, the background noise component Is determined for each predetermined time unit, that is, for each frame, that is, the second state mainly including the first state and the third state other than the first and second states including both the voice component and the background noise component. (Step S100
2).

Then, based on the determination result, the selecting unit 1
In step 05, the main component of the voice, which is the first component, and the main component of the background noise, which is the second component, are selected. That is, when the input audio signal is in the first state, the input audio signal is output as it is as an audio main component, and a predetermined signal, for example, an all-zero signal S3 is output as a background noise main component (step S1003). If the input audio signal is in the second state, a predetermined signal, for example, an all-zero signal S3, is output as the main audio component, and the input audio signal is output as it is as the main background noise component (step S1004). further,
When the input audio signal is in the third state, the first audio component extraction signal S1 is output as the main audio component, and the noise component extraction signal S3 is output as the main background noise component (step S1005). The audio main component and the background noise main component output in this way are encoded by an audio encoding unit and a noise encoding unit, respectively, as described later.

As described above, according to the present embodiment, when the input speech signal is in the third state including both the speech component and the background noise component, the first noise suppression processing unit 101 having a relatively small suppression amount performs Since the first audio component extraction signal S1 is output and is encoded by the audio encoding unit using an encoding method suitable for audio, it is possible to avoid quality degradation due to loss of information of audio components.

The flowchart of FIG.
0 shows another processing procedure, and is the same as FIG. 6 except that step S1004 in FIG. 6 is changed to step S1006. In this step S1006,
When the input audio signal is in the second state, the second audio component extraction signal S2 is output as the main audio component, and the noise component extraction signal S3 is output as the background noise main component. Are almost the same.

Next, the processing procedure of the state determination unit 104 shown in FIG. 5 will be described with reference to the flowchart shown in FIG.
This process is also performed for each frame of the input audio signal. First, the estimated energy Eest of the noise component extraction signal S3 is calculated by the following equation (1) (step S1).
101). Eest (m) = α · Eno + (1−α) · Eest (m−1) (1) Here, α represents an update coefficient, and m represents a frame number of an input audio signal. It is calculated according to E _no (3).

Next, the SNR (S / N ratio) of the second audio component extraction signal S2 and the noise component extraction signal S3, that is, both signals S
The energy ratio between S2 and S3 is calculated as follows (step S1102). First, the energy E _{sp of} the second audio component extraction signal S2 and the energy E _no of the noise component extraction signal S3 are calculated by the following equations (2) and (3).

[0042]

(Equation 1)

Here, sp (n) represents the second audio component extraction signal S2, no (n) represents the noise component extraction signal S3,
LN represents the frame length (= 160). Since the calculated energy E _{sp of} the audio component extraction signal S2 and the energy E _no of the noise component extraction signal S3 are signals in the log region, the difference E _sp −E _no between them is the energy ratio of the signals S2 and S3. It will represent the SNR.

Next, compared with the threshold TH1 as the SNR (= E _sp -E _no) the following equation (4) (step S11
03). Where _{SNR (= E sp -E no)} ≧ TH1 (4), in the case of satisfying the formula (4), the input speech signal is a candidate for audio-oriented components. On the other hand, if the expression (4) is not satisfied, the pitch analysis is performed on the noise component extraction signal S3, and the magnitude G _no of the pitch periodicity is calculated by the following expressions (5) and (6) (step S1104).

[0045]

(Equation 2)

Here, TMIN is the minimum value of the pitch period,
TMAX represents the maximum value of the pitch period, respectively, and NAN
A represents the pitch analysis length.

Next, the magnitude G _no of the pitch periodicity calculated in this manner is compared with a threshold value TH2 as shown in the following equation (7) (step S1105). G _no ≧ TH2 (7) Here, when Expression (7) is satisfied, the first state is selected assuming that the input audio signal is a signal mainly including an audio component (Step S1107), and Expression (7) is obtained. If not, the second state is selected assuming that the input audio signal is a signal mainly composed of noise components (step S110).
8).

On the other hand, if it is determined in step S1103 that expression (4) is satisfied and the input audio signal is a candidate for the main audio component, then in step S1106, the noise component extraction signal calculated by expression (1) The estimated energy Eest of S3 is compared with a threshold value TH3 as in the following equation (8). Eest (m) ≧ TH3 (8) Here, when Expression (8) is satisfied, the input voice signal has a relatively large voice component and a relatively large background noise component. The third state is selected as including both (step S110)
9). On the other hand, if Expression (8) is not satisfied, it can be determined that the background noise component in the input audio signal can be ignored.
The first state is selected (step S1107).

As described above, when determining the state of the input speech signal, the magnitude G _no of the pitch periodicity of the noise component extraction signal S2 is determined in steps S1104 to S1105, and it is checked whether the magnitude G _no is equal to or larger than the threshold TH2. , G _no is the threshold T
In the case of H2 or more, the second state is not selected, the first state is selected and treated in the same manner as the audio main component, and the audio is encoded by the audio encoding unit. The noise component extraction signal S3 having a large periodicity in the encoding unit
Can be avoided. Note that the threshold values TH1, TH2, T
H3 may be a predetermined fixed value or a time-varying value that is adaptively required to further improve performance.

The signal used in state determination section 104 is
Here, the description has been given using the second audio component extraction signal S2 and the noise component extraction signal S3, but the combination of the first audio component extraction signal S1 and the noise component extraction signal S3 or the first audio component extraction signal S1 and the input signal First audio component extraction signal S
Similar processing is performed using a combination of a signal obtained by subtracting 1 or a combination of the second audio component extraction signal S2 and a signal obtained by subtracting the first audio component extraction signal S1 from the input audio signal. It is possible.

(Second Embodiment) FIG. 9 shows a second embodiment of the present invention.
1 shows a configuration of a component separating unit 100 to which a component separating method of an audio signal according to an embodiment is applied. The component separating unit 100 includes a frequency domain transforming unit 201, a noise spectrum estimating unit 202,
The first gain vector calculation unit 203, the second gain vector calculation unit 204, the multipliers 205 and 206, the time domain conversion units 207 and 208, the switch 209, the update unit 210, the adder 103, the state determination unit 104, and the selection unit 105 Be composed. In FIG. 9, the same components as those in FIG. 4 are denoted by the same reference numerals, and detailed description will be omitted.

The features of this embodiment include two noise suppression processes having different amounts of suppression as in the first embodiment, and a part of these two noise suppression processes in order to reduce the complexity of the process. Is common. Hereinafter, FIG.
Using the flowchart shown in FIG.
Will be described. This process is performed for each frame of the input audio signal. Although the case where the spectrum subtraction method is used for the noise suppression processing in the present embodiment will be described, it is also possible to realize the noise suppression processing using another method.

First, the input voice signal, which is a signal in the time domain, is converted to the frequency domain by the frequency domain conversion section 201 to obtain a spectrum coefficient (step S2101). Next, a first gain vector is calculated by a first gain vector calculator 203 having a relatively small amount of suppression based on an estimated spectrum of background noise estimated from past background noise by a noise spectrum estimator 202, Then, the second gain vector is calculated by the second gain vector calculation unit 204 having a large suppression amount (step S2102). Here, the first gain vector and the second gain vector refer to a plurality of gains that are multiplied by the corresponding spectral coefficients of the input audio signal, and the expression “vector” is used for that purpose. That is,
In step S2102, a corresponding gain is determined for each spectral coefficient of the input audio signal. Further, instead of each spectral coefficient, a band may be divided into a predetermined bandwidth and a gain to be suppressed for each band may be obtained. In this case, the gain for the number of bands is calculated.

Next, the multiplier 205 multiplies the spectral coefficient of the input audio signal by the gain corresponding to each of the first gain vectors (step S2103), and the time domain inverse transform section 207 performs inverse transform on the time domain to obtain the first speech signal. Component extraction signal S
1 is generated (step S2104). Similarly, the multiplier 206 multiplies the spectral coefficient of the input audio signal by the corresponding gain of the second gain vector (step S2).
105), and inversely transform to the time domain by the time domain inverse transform unit 208 to generate the second audio component extraction signal S2 (Step S)
2106).

Alternatively, the audio component signal may be extracted by subtracting the estimated value of the noise spectrum from the spectral coefficient of the input audio signal. in this case,
The first gain vector calculation unit 203 and the second gain vector calculation unit 204 each have a function of calculating a subtraction value,
Further, the multipliers 205 and 206 are replaced with subtractors.

Next, the noise component extraction signal S3 is generated by subtracting the second audio component extraction signal S2 from the input audio signal by the subtractor 103 (step S2107).

Next, the state of the input audio signal, that is, the first state in which the input audio signal mainly includes the audio component, the background noise component is obtained from the second audio component extraction signal S2 and the noise component extraction signal S3 by the state determination unit 104. Is determined for each predetermined time unit, that is, for each frame, that is, the second state mainly based on, and the third state including both the voice component and the background noise component (step S2108).

Then, based on this determination result, the selecting unit 1
At 05, a main component of speech and a main component of background noise are selected.
That is, when the input audio signal is in the first state, the input audio signal is output as it is as the audio main component, and the all-zero signal S4 is output as the background noise main component (step S2109). If the input audio signal is in the second state, the all-zero signal S4 is output as the main audio component, and the input audio signal is output as it is as the main background noise component (step S2110). Further, if the input audio signal is
In the case of (1), the first audio component extraction signal S1 is output as the main audio component, and the noise component extraction signal S3 is output as the main background noise component (step S2111). The audio main component and the background noise main component output in this manner are encoded by an audio encoding unit and a noise encoding unit (not shown), respectively.

Next, it is determined whether or not to update the estimated value of the noise spectrum for the processing of the input voice signal of the next frame (step S2113). However, basically, the method of updating the estimated value of the background noise is not directly related to the gist of the present invention, and an existing method may be applied. For example,
Since it is considered that the estimated value of the noise spectrum must always reflect the latest spectrum shape of the background noise, the second component that can be regarded as having the main component of the background noise in the current frame is considered.
When the third state and the third state are selected, the switch 209 is turned on to activate the updating unit 210. When the first state in which the background noise main component does not exist is selected, the switch 209 is turned off so that the updating unit 210 is not activated. In the updating unit 210, the switch 209
Is operated only when is turned on, and updating is performed so that the spectral coefficient obtained in the current frame is reflected in the estimated value of the noise background (step S2114).

As described above, according to the present embodiment, the processing complexity is reduced as compared with the first embodiment, and when the input speech signal is in the third state including both the speech component and the background noise component. Is output from a noise suppression processing unit with a small amount of suppression, and is encoded by an encoding method suitable for audio in an audio encoding unit, thereby avoiding quality degradation due to information loss of audio components. be able to.

The flowchart shown in FIG. 12 shows another processing procedure of the component separation unit 100 shown in FIG. 9. The processing shown in FIGS. 10 to 11 is performed except that step S2110 in FIG. 11 is changed to step S2112. The procedure is the same. In step S2112, when the input audio signal is in the second state, the second audio component extraction signal S2 is output as the main audio component, and the noise component extraction signal S3 is output as the main background noise component. Even so, the result is almost the same.

(Third Embodiment) Next, a speech encoding / decoding method using a speech signal component separation method according to the present invention will be described as a third embodiment of the present invention.

[On the Encoding Side] FIG. 13 shows the configuration of a speech encoding apparatus to which the speech encoding method according to the third embodiment of the present invention is applied. This speech coding apparatus includes a component separation unit 100, a bit allocation selection unit 120, a speech coding unit 130, a noise coding unit 140, and a multiplexing unit 150.

The component separation unit 100 analyzes the input audio signal for each predetermined time unit, and determines a first component (main audio component) mainly composed of audio and a second component mainly composed of background noise. (The main component of the background noise), and has the same configuration as that described in the first embodiment.

The bit allocation selection unit 120 determines whether the speech encoding unit 130 and the background noise encoding unit 140
Are selected from a predetermined combination of bit allocations, and information on those bit allocations is output to speech coding section 130 and noise coding section 140. At the same time, information on the result of the determination by the state determination unit 104 is output to the multiplexing unit 150 as transmission information.

This bit allocation is performed by the state determination unit 104
It is desirable to select according to the determination result in, but is not limited thereto.For example, in order to obtain a more stable sound quality, a sudden change in the bit allocation is unlikely to occur while monitoring the change in the bit allocation from the past. It is also effective to determine the bit allocation by combining the mechanisms. As an example of combinations of bit allocation prepared in the bit allocation selection unit 120 and codes representing the combinations, those shown in the following Table 1 can be considered.

[0067]

[Table 1]

According to Table 1, when the mode "0" is selected, 78 bits per frame are allocated to the voice coding unit 130, and no bit is allocated to the noise coding unit 140. Since 2 bits of the bit allocation code are sent to this, the total number of bits required for encoding the input audio signal is 80
Bit. It is desirable to select this mode “0” bit allocation for a frame in which there is almost no background noise main component as compared to the voice main component. By doing so, the bit allocation to the audio encoding unit is increased, and the sound quality of the reproduced audio is improved.

On the other hand, when the mode “1” is selected, no bit is allocated to the voice coding unit 130, and 78 bits are allocated to the noise coding unit 140. Since two bits of the bit allocation code are sent to this, the total number of bits required for encoding the input audio signal is 80 bits. It is desirable to select the mode “2” bit allocation in a frame in which the speech main component can be ignored relative to the background noise main component.

When mode “2” is selected, 78-Y bits are assigned to speech encoding section 130, and Y bits are assigned to noise encoding section 140. Here, Y represents a sufficiently small positive integer. Here Y = 8
, But is not limited to this value. In mode "2", in addition to this, two bits of the bit allocation code are sent, so that the total number of bits required for encoding the input audio signal is 80 bits.

It is desirable to select a bit assignment such as in mode "2" for a frame in which both a speech main component and a background noise main component coexist. In this case, since the audio main component is obviously more important perceptually, the number of bits allocated to the noise encoding unit 140 is extremely small, and the number of bits allocated to the audio encoding unit 130 is increased by the reduced amount. Thus, the main audio component can be accurately encoded. At this time, the point is how efficiently the main component of the background noise can be encoded with a small number of bits in the noise encoding unit 140. The specific realization method is described in, for example, Japanese Patent Application No. 10-128876. The described technique may be used.

In this manner, speech and background noise can be efficiently encoded by the respective encoding units.
Sound with natural background noise can be reproduced.
A suitable frame length is about 10 to 30 ms in speech coding. In this example, the total number of bits per frame is fixed at 80 for three combinations of bit allocation. As described above, when the total number of bits per frame is fixed, encoding can be performed at a fixed bit rate regardless of the input voice.

Although three types of bit allocation are shown here, it is apparent that the present invention can be applied to a configuration using still another type of bit allocation.

[0074] The speech encoding unit 130
The audio signal from inputs a main component, and the main component is an audio signal by audio coding that reflects the characteristics of the audio signal. It goes without saying that any encoding method may be used for the audio encoding unit 130 as long as the encoding method can efficiently encode the audio signal. However, as an example, a more natural encoding method is used here. As a method capable of providing voice, a CELP scheme will be used. The CELP method is a method of performing coding in a normal time domain, and is characterized by performing coding of an excitation signal such that distortion of a synthesized waveform in the time domain is reduced.

The noise encoding section 140 includes the component separating section 100
Is input so that the main component of the background noise can be input, and the noise can be appropriately encoded. Normally, the background noise signal has a gradual temporal change in spectrum as compared with the speech signal. Further, the phase information of the waveform is also almost random, and the phase information is not so important for the human ear. In order to encode such background noise components efficiently, rather than transforming waveforms such as the CELP method to reduce waveform distortion, a transform from a time domain to a transform domain is performed as in transform coding. The method of encoding the parameters extracted from the coefficients or the transform coefficients enables more efficient encoding. In particular, when encoding is performed in consideration of the characteristics of human auditory sense after conversion into the frequency domain, the encoding efficiency can be further increased.

Next, the processing procedure of the speech encoding method of this embodiment will be described with reference to the flowchart of FIG. First, an input audio signal is fetched every predetermined time unit (step S100), and the component separation unit 100 analyzes the input audio signal and separates the input audio signal into a main audio component and a main background noise component (step S101).

Next, the number of bits to be allocated to each of speech coding section 130 and background noise coding section 140 is determined in advance by bit allocation selecting section 120 based on the determination result of state determining section 103 in component separating section 100. In addition to selecting from the combinations of the allocated numbers of bits, the information of the determination result is output to the speech encoding unit 130 and the background noise encoding unit 140 (step S102).

Then, according to the respective bit allocations selected by the bit allocation selecting section 120, the coding processing is performed by the voice coding section 130 and the noise coding section 140 (step S103). Specifically, the audio encoding unit 130 receives the audio main component from the component separation unit 100, performs encoding with the number of bits allocated to the audio encoding unit 130, and Find coded data.

On the other hand, the noise encoding section 140 receives a component mainly composed of the background noise signal from the component separating section 100, encodes the noise with the number of bits allocated to the noise encoding section 140, and reduces the background noise. The encoded data for the main component is obtained.

Next, in the multiplexing section 150, each of the encoding sections 13
0, 140 and each of the encoding units 130, 1
The information of the bit allocation to 40 is multiplexed and output as transmission coded data to the transmission path (step S104).
This completes the encoding process performed within the predetermined time interval. Then, it is determined whether to continue or end the encoding for the next time section (step S105).

In the above embodiment, the component separating section 100
When the input audio signal is subjected to component separation into a main component of a voice and a main component of a background noise, if the separation performance is poor, the quality of the decoded voice signal deteriorates. For example, if noise is suppressed by passing an input audio signal through a noise canceller to perform this component separation, even if there is no background noise in the input audio signal, distortion is generated in the main audio component by unnecessary noise suppression operation. When this is input to the audio encoding unit 130, the quality of the decoded audio signal is degraded.

On the other hand, background noise does not usually have periodicity. For example, in background noise where a person is talking at a distance,
There is periodicity. In such a case, if the periodicity remains in the background noise main component obtained after the component separation, and this is input to the noise encoding unit 140 as it is, unlike the speech encoding unit 130, the noise encoding unit 140 In addition, because the periodicity (pitch period) cannot be expressed efficiently, the background noise main component having strong pitch periodicity (speaker interference,
For example, bubble noise).

On the other hand, if the component separation unit 100 uses the component separation method of the audio signal described in the first embodiment,
The components can be accurately separated without such a problem.

[Specific Example of Speech Encoding Apparatus] FIG.
2 shows a specific example of a speech encoding apparatus in the case of using transform coding. In the CELP method, a vocal cord signal is made to correspond to a sound source signal as a model of a sound generation process, a spectral envelope characteristic represented by a vocal tract is represented by a synthesis filter, and the sound source signal is input to the synthesis filter. To express. At this time, the feature is that the excitation signal is encoded so that the waveform distortion between the audio signal provided for CELP encoding and the audio reproduced after encoding is perceptually reduced.

Speech encoding section 130 receives the speech main component from component separating section 100 and encodes this component such that waveform distortion in the time domain is reduced. At this time, the coding unit 130 is allocated under the assignment of a predetermined number of bits according to the bit assignment from the bit assignment selecting unit 120.
Are performed. In this case, the sum of the number of bits used in each encoding unit in encoding unit 130 is made to match the bit allocation from bit allocation selecting unit 120 to encoding unit 130, so that the performance of speech encoding unit 130 is maximized. It can be used for This means that the encoding unit 14
The same applies to 0.

The CELP coding described here includes a spectrum envelope codebook search section 311 and an adaptive codebook search section 31.
2, noise codebook search section 313, gain codebook search section 31
4 is encoded. Codebook search units 311 to 31
The information of the index of the codebook searched in 4 is input to the encoded data output unit 315, and is output from the encoded data output unit 315 to the multiplexing unit 150 as audio encoded data.

Next, the function of each of the codebook search sections 311 to 314 in the voice coding section 130 will be described. The spectrum envelope codebook search unit 311 includes the component separation unit 100
From the speech-based component for each frame, search for a spectral envelope codebook prepared in advance, and select an index of a codebook that can better express the spectral envelope of the input signal, The information of the index is output to the encoded data output unit 315. Usually CE
In the LP method, an LSP (Line Spectrum Pair) parameter is used as a parameter used when encoding a spectrum envelope. However, the present invention is not limited to this, and other parameters are also effective as long as they can express a spectrum envelope.

The adaptive codebook search section 312 is used to express a component that repeats at a pitch period in the sound source.
In the CELP system, the encoded past excitation signal is stored for a predetermined length as an adaptive codebook, and is stored in both the audio encoding unit and the audio decoding unit, so that the signal corresponds to the specified pitch period. And a signal to be repeated is extracted from the adaptive codebook. In the adaptive codebook, since the output signal from the codebook and the pitch cycle correspond one-to-one, the pitch cycle can be made to correspond to the index of the adaptive codebook. Under such a structure, the adaptive codebook search unit 312 evaluates the distortion between the synthesized signal obtained when the output signal from the codebook is synthesized by the synthesis filter and the target audio signal at a level weighted by auditory perception. Search for an index of the pitch period such that is smaller. Then, the information of the searched index is output to the encoded data output unit 315.

The noise codebook search section 313 is used to represent a noise component in the sound source. In the CELP system, a noise component of a sound source is represented using a noise codebook, and has a structure capable of extracting various noise signals from the noise codebook in accordance with a specified noise index. Under such a structure, the noise codebook search unit 313 perceptually weights the distortion between the synthesized speech signal reproduced using the output signal from the codebook and the target speech signal in the noise codebook search unit 313. Evaluate at the level and search for a noise index that reduces the distortion. Then, the information of the searched noise index is output to encoded data output section 315.

Gain codebook search section 314 is used to represent the gain component of the excitation. In the CELP system, two kinds of gains, a gain used for a pitch component and a gain used for a noise component, are encoded by a gain codebook search unit 314. In the codebook search, the distortion between the synthesized speech signal reproduced using the gain candidates extracted from the codebook and the target speech signal is evaluated at an auditory weighted level, and a gain index that reduces the distortion is evaluated. Explore. Then, the searched gain index is output to encoded data output section 315. Encoded data part 3
15 outputs the encoded data to the multiplexing unit 150.

Next, the background noise main component from the component separation unit 100 is inputted, and the noise
A detailed configuration example will be described with reference to FIG.

The noise encoding unit 140 includes the component separating unit 100
The above-described speech encoding unit 130 is characterized in that the background noise main component from the input is input, a transform coefficient of this component is obtained using a predetermined transform, and encoding is performed so that parameter distortion in the transform domain is reduced. It is very different from the encoding method used. Various methods are conceivable for expressing parameters in the transform domain.Here, as an example, a background noise component is divided into bands in the transform domain, a parameter representative of the band is obtained for each band, and the parameter is set to a predetermined value. A method of quantizing with a quantizer and transmitting the index will be described.

First, the transform coefficient calculating section 321 obtains the transform coefficient of the main component of the background noise by using a predetermined transform.
As a method of conversion, for example, discrete Fourier transform or FFT
(Fast Fourier transform) can be used. Next, the band dividing section 322 divides the frequency axis into predetermined bands, and uses the number of quantization bits according to the bit allocation from the noise coded bit allocation section 320 to generate the first band coding section 323 and the second band coding section 323. The band coding units 324,..., And the m-th band coding unit 325 perform parameter quantization for each of the m bands. In 8 kHz sampling, the number m of bands is preferably a value of about 4 to 16.

As the parameter at this time, a value obtained by averaging the spectrum amplitude and the power spectrum obtained from the transform coefficients in each band can be used.
Index information indicating the quantization value of the parameter from each band is input to the encoded data output unit 326, and is output from the encoded data output unit 326 to the multiplexing unit 150 as encoded data.

In the present embodiment, the main component of the background noise is generated by subtracting the output of the noise suppression processing unit having a large amount of suppression from the input speech signal, and this is converted into the noise encoding unit 1.
Although the description has been given of the configuration in which the input is 40, the estimated background noise spectrum held in the component separating section 100 is processed and directly supplied to the noise encoding section 140, and the spectrum is divided into bands to code the respective bands. You may make it. In this case, the signal processing is substantially equivalent to the above embodiment, and the processing result can be realized without change, and an extra conversion process can be omitted, so that the effect of reducing the processing amount can be obtained.

[On the Decoding Side] FIG. 16 shows a configuration of a speech decoding apparatus to which the speech decoding method according to the present embodiment is applied. This speech decoding device includes a demultiplexing section 160, a bit allocation decoding section 170, a speech decoding section 180, a noise decoding section 190, and a combining section 195.

The demultiplexing section 160 receives the transmission coded data transmitted from the voice coding apparatus shown in FIG. 13 every predetermined time unit as described above, and uses this data for bit allocation information and voice decoding. Coded data to be input to the decoding section 180 and coded data to be input to the noise decoding section 190 and output.

Bit allocation decoding section 170 decodes the information of bit allocation, and allocates the number of bits to each of speech decoding section 180 and noise decoding section 190 to the number of bits determined by the same mechanism as that of the encoding side. Output from the combination of.

Speech decoding section 180 decodes the encoded data based on the bit assignment by bit assignment decoding section 170 to generate a reproduced signal for a component mainly composed of speech, and outputs this to combining section 195. .

The noise decoding section 190 decodes the encoded data based on the bit allocation by the bit allocation decoding section 170 to generate a reproduced signal for a component mainly composed of background noise, and outputs this to the combining section 195. I do.

The combining section 195 combines the reproduced signal mainly composed of the sound decoded and reproduced by the audio decoding section 180 with the reproduced signal mainly composed of the noise decoded and reproduced by the noise decoding section 190. Then, a final output audio signal is generated.

Next, the processing procedure of the speech decoding method of this embodiment will be described with reference to the flowchart of FIG. First, the input transmission coded data is fetched for each predetermined time unit (step S200), and the coded data is demultiplexed by the demultiplexing section 160 to obtain bit allocation information and a code to be input to the speech decoding section 180. It is separated into encoded data and encoded data to be input to the noise decoding unit 190 and output (step S201).

Next, bit allocation information is decoded in bit allocation decoding section 170 and speech decoding section 180 is decoded.
And the number of bits to be allocated to each of the noise decoding unit 190 is output from a combination of bit number allocations determined by the same mechanism as that of the speech coding apparatus (step S2).
02). Then, based on the bit allocation from bit allocation decoding section 170, speech decoding section 180 and noise decoding section 190 generate respective reproduction signals from the respective coded data and output these to combining section 195 (step S203).

Next, the combining section 195 combines the component mainly composed of the reproduced audio signal and the reproduced component mainly composed of the noise signal (step S204), and generates and outputs a final audio signal. (Step S205).

[Specific Example of Speech Decoding Apparatus] FIG. 18 shows a specific example of a speech decoding apparatus corresponding to the configuration of the speech coding apparatus of FIG. The demultiplexing unit 160 converts the transmission coded data for each predetermined time unit transmitted from the voice coding apparatus of FIG.
In addition to outputting the spectral envelope index, the adaptive index, the noise index, and the gain index information which are the encoded data to be input to 0, each band which is the encoded data to be input to the noise decoding unit 190 Outputs quantization index information. The bit allocation decoding unit 170 decodes the information of the bit allocation, and determines the number of bits to be allocated to each of the speech decoding unit 180 and the noise decoding unit 190 in a combination of the allocation of the number of bits determined by the same mechanism as the encoding. Output from

Speech decoding section 180 decodes the coded data based on the bit assignment from bit assignment decoding section 170 to generate a reproduced signal for a component mainly composed of speech, and outputs this to combining section 195. . In particular,
The spectrum envelope decoding unit 414 reproduces information of the spectrum envelope from the spectrum envelope index and the spectrum envelope codebook prepared in advance, and reproduces the information of the spectrum envelope.
Send to 6. Further, adaptive excitation decoding section 411 inputs adaptive index information, extracts a signal repeated at a corresponding pitch cycle from the adaptive codebook, and generates a signal from excitation excitation reproducing section 415.
Output to

The noise excitation decoding section 412 receives the information of the noise index, extracts a noise signal corresponding to the information from the noise codebook, and outputs it to the excitation reproducing section 415.

Gain decoding section 413 receives the information of the gain index, extracts two kinds of gains corresponding to the gain index, ie, a gain used for the pitch component and a gain used for the noise component, from the gain codebook and outputs them to sound source reproducing section 415. I do.

The sound source reproducing unit 415 is adapted to
A signal (vector) Ep that repeats at a pitch cycle from 1
And a noise signal (vector) from the noise excitation decoding unit 412
En and two types of gains G from the gain decoding unit 413
The sound source signal (vector) Ex is reproduced using p and Gn according to the following equation (1). Ex = Gp · Ep + Gn · En (9) The synthesis filter 416 sets the parameters of the synthesis filter for synthesizing the voice using the information of the spectrum envelope, and synthesizes by inputting the sound source signal from the sound source reproduction unit 415. Generate an audio signal. Further, the post-filter 417 shapes the coding distortion included in the synthesized voice signal and outputs the synthesized voice signal to the combining unit 195 so as to make the sound easy to hear.

Next, the noise decoding section 190 in FIG. 18 will be described. The noise decoding unit 190 determines whether the noise decoding unit 1
The necessary coded data is input to 90, which is decoded to generate a reproduced signal of a component mainly composed of background noise, and outputs the reproduced signal to the combining unit 195. Specifically, the coded data is separated into quantization indices for each band by the noise data separation unit 420, and the first band decoding unit 421, the second band decoding unit 422, ..., the m-th band decoding unit According to 423, the parameters in each band are decoded, and the inverse of the conversion performed on the encoding side is performed using the parameters decoded by the inverse transform unit 424, and a component mainly composed of background noise is converted into a reproduced signal. Generate. The reproduced signal mainly composed of the background noise is
It is sent to the coupling unit 195.

The combining unit 195 combines the components mainly composed of the voice shaped by the post-filter and the reproduced signals mainly composed of the background noise so as to be smoothly connected between adjacent frames. Then, this is used as an output audio signal as the final output from the decoding unit.

(Fourth Embodiment) FIG. 19 shows a fourth embodiment of the present invention using the component separation unit 100 described in the second embodiment.
1 shows a configuration of a speech encoding device according to an embodiment.
The speech coding apparatus according to the present embodiment is obtained by replacing the configuration of the component separation unit 100 in the third embodiment with the component separation unit 100 described in the second embodiment. Since it is clear from the third embodiment, the same parts as those in the second and third embodiments are denoted by the same reference numerals, and detailed description is omitted.

Note that the method for separating components of an audio signal according to the present invention can be applied to uses other than audio coding.
For example, the present invention can be widely applied to uses in which different processing is performed on both components, such as separately recording a main component of speech and a main component of background noise.

[0114]

As described above, according to the present invention, there is provided an audio signal component separation method capable of accurately separating an audio main component and a background noise main component from an input audio signal. Can be.

Further, according to the present invention, there is provided a low-rate speech encoding method capable of reproducing speech including background noise as close as possible to the original sound by using such a component separation method. Can be.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining a problem in a case where noise suppression processing with a large amount of suppression is used for separating components of an audio signal.

FIG. 2 is a diagram for explaining a problem in a case where noise suppression processing with a small amount of suppression is used to separate components of an audio signal.

FIG. 3 is a diagram for explaining the principle of component separation of an audio signal according to the present invention;

FIG. 4 is a block diagram illustrating a configuration of a component separation unit according to the first embodiment of the present invention.

FIG. 5 is a block diagram illustrating a configuration of a state determination unit in FIG. 1;

FIG. 6 is a flowchart illustrating an example of a processing procedure of a component separation unit according to the first embodiment.

FIG. 7 is a flowchart illustrating another example of the processing procedure of the component separation unit according to the first embodiment.

FIG. 8 is a flowchart showing a processing procedure of a state determination unit in FIG. 5;

FIG. 9 is a block diagram illustrating a configuration of a component separation unit according to a second embodiment of the present invention.

FIG. 10 is a block diagram illustrating a part of a processing procedure of a component separation unit according to the second embodiment.

FIG. 11 is a block diagram illustrating another part of the processing procedure of the component separation unit according to the second embodiment.

FIG. 12 is a block diagram illustrating another example of the processing procedure of the component separation unit according to the second embodiment.

FIG. 13 is a block diagram illustrating a schematic configuration of a speech encoding device according to a third embodiment of the present invention.

FIG. 14 is a flowchart illustrating a processing procedure of the speech encoding device according to the third embodiment;

FIG. 15 is a block diagram showing a more detailed configuration of the speech coding apparatus according to the third embodiment;

FIG. 16 is a block diagram showing a schematic configuration of a speech decoding device according to a third embodiment.

FIG. 17 is a flowchart showing a processing procedure of a speech decoding method according to the third embodiment;

FIG. 18 is a block diagram showing a more detailed configuration of the speech decoding device according to the third embodiment.

FIG. 19 is a block diagram showing a configuration of a speech encoding device according to a fourth embodiment of the present invention.

[Explanation of symbols]

 Reference Signs List 100 component separation unit 101 first noise suppression processing unit 102 second noise suppression processing unit 103 subtractor 104 state determination unit 105 selection unit 111 SNR determination unit 112 periodicity determination unit 113 estimated energy determination Unit 120: bit allocation selecting unit 130: voice coding unit 140: noise coding unit 150: multiplexing unit 160: demultiplexing unit 170: bit allocation decoding unit 180: voice decoding unit 190: noise decoding unit 195 Joint

Claims (7)

[Claims] 1. A method of separating an input audio signal into a first component mainly composed of voice and a second component mainly composed of background noise, for each predetermined time unit, comprising: In contrast, a signal obtained by performing a first noise suppression process with a relatively small amount of suppression is used as the first component, and a second noise suppression process with a relatively large amount of suppression is performed on the input speech signal. A method for separating components of an audio signal, comprising a separation mode in which a signal obtained by subtracting a signal obtained from the input audio signal is used as the second component. 2. A method for separating an input audio signal into a first component mainly composed of voice and a second component mainly composed of background noise, for each predetermined time unit, comprising: In contrast, the first signal obtained by performing the first noise suppression processing having a relatively small amount of suppression and the second signal obtained by performing the second noise suppression processing having a relatively large amount of suppression on the input voice signal. A second signal, and a third signal obtained by subtracting the second signal from the input audio signal, wherein the input audio signal includes a first state mainly including an audio component, and a background noise component. Determining, for each of the predetermined time units, a main second state and a third state other than the first and second states including both a voice component and a background noise component; When the audio signal is determined to be in the first state, the first component And outputting the input audio signal as it is, and outputting a predetermined signal as the second component. When the input audio signal is determined to be in the second state, the first A predetermined signal is output as a component, and the input audio signal is output as the second component. When the input audio signal is determined to be in the third state, the first component is output. Outputting the first signal and outputting the third signal as the second component. 3. A method for separating an input audio signal into a first component mainly composed of voice and a second component mainly composed of background noise, for each predetermined time unit, comprising: In contrast, the first signal obtained by performing the first noise suppression processing having a relatively small amount of suppression and the second signal obtained by performing the second noise suppression processing having a relatively large amount of suppression on the input voice signal. A second signal, and a third signal obtained by subtracting the second signal from the input audio signal, wherein the input audio signal includes a first state mainly including an audio component, and a background noise component. Determining, for each of the predetermined time units, a main second state and a third state other than the first and second states including both a voice component and a background noise component; When the audio signal is determined to be in the first state, the first component And outputting the input audio signal as it is, and outputting a predetermined signal as the second component. When the input audio signal is determined to be in the second state, the first Outputting the second signal as a component, outputting the third signal as the second component, and when the input audio signal is determined to be in the third state, as the first component A method for separating components of an audio signal, comprising outputting the first signal and outputting the third signal as the second component. 4. The method according to claim 1, wherein said input audio signal is in said first state,
4. The voice signal according to claim 2, wherein the process of determining which of the third state and the third state includes a step of checking a magnitude of the pitch periodicity of the third signal. Component separation method. 5. An input audio signal is separated into a first component mainly composed of voice and a second component mainly composed of background noise for each predetermined time unit, and based on the first and second components. Alternatively, based on the determination result of the state determined at the time of the component separation, the bit allocation of each component is selected from a plurality of predetermined bit allocation candidates. And a second component are respectively encoded by different predetermined encoding methods, and the encoded data of the first and second components and the information of the bit allocation are output as transmission encoded data. In the component separation, a signal obtained by performing a first noise suppression process with a relatively small amount of suppression on the input audio signal is defined as the first component, and the relative amount of suppression with respect to the input audio signal is determined. Large second speech encoding method characterized by having a separation mode in which a signal obtained by subtracting from the input speech signal a signal obtained by performing a noise suppressing process to the second component. 6. An input audio signal is separated into a first component mainly composed of voice and a second component mainly composed of background noise for each predetermined time unit, and based on the first and second components. Alternatively, based on the determination result of the state determined at the time of the component separation, the bit allocation of each component is selected from a plurality of predetermined bit allocation candidates. And a second component are respectively encoded by different predetermined encoding methods, and the encoded data of the first and second components and the information of the bit allocation are output as transmission encoded data. In the component separation, a first signal obtained by performing a first noise suppression process with a relatively small suppression amount on the input speech signal, a first signal with a relatively large suppression amount on the input speech signal, 2 Generating a second signal obtained by performing the noise suppression processing and a third signal obtained by subtracting the second signal from the input audio signal, wherein the input audio signal mainly includes an audio component; The predetermined state, the first state, the second state mainly including the background noise component, and the third state other than the first and second states including both the voice component and the background noise component. When the input voice signal is determined to be in the first state, the input voice signal is output as it is as the first component, and a predetermined value is determined as the second component. When the input voice signal is determined to be in the second state, a predetermined signal is output as the first component, and the input voice signal is output as the second component. Output a signal, and the input audio signal is When it is determined that the signal is in the third state, the first signal is output as the first component, and the third signal is output as the second component. Method. 7. An input audio signal is separated into a first component mainly composed of voice and a second component mainly composed of background noise every predetermined time unit, and based on the first and second components. Alternatively, based on the determination result of the state determined at the time of the component separation, the bit allocation of each component is selected from a plurality of predetermined bit allocation candidates. And a second component are respectively encoded by different predetermined encoding methods, and the encoded data of the first and second components and the information of the bit allocation are output as transmission encoded data. In the component separation, a first signal obtained by performing a first noise suppression process with a relatively small suppression amount on the input speech signal, a first signal with a relatively large suppression amount on the input speech signal, 2 Generating a second signal obtained by performing the noise suppression processing and a third signal obtained by subtracting the second signal from the input audio signal, wherein the input audio signal mainly includes an audio component; The predetermined state, the first state, the second state mainly including the background noise component, and the third state other than the first and second states including both the voice component and the background noise component. When the input voice signal is determined to be in the first state, the input voice signal is output as it is as the first component, and a predetermined value is determined as the second component. When the input audio signal is determined to be in the second state, the second signal is output as the first component, and the third signal is output as the second component. And the input audio signal is in the third state And when it is determined, the first outputs the first signal as a component, the speech encoding method and outputting the third signal as the second component.