JPH10143195A - Post filter - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the post filter in which reproduced sounds having no sense of incongruity are obtained even in the consonant section, that has no pitch constitution nor formant peaks, and in a no voice segment. SOLUTION: If the input signals are voice signals, a selection is made to execute a pitch control filter 3 by a selection switch 2 and also executions are made for a formant control filter 4 and a spectrum inclination compensation filter 5. If the input signals are no voice signals, the execution of the filter 3 is prohibited by the switch 2, a coefficient controlling section 7 controls the formant of the filter 4 so that the formant becomes flat and the inclination of the spectrum of the filter 5 is maintained.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to a post filter,
More specifically, the present invention relates to a post filter used when decoding audio data obtained by encoding an audio signal with high efficiency.

[0002]

2. Description of the Related Art Conventionally, as means widely used for efficiently compressing a speech signal, encoding is performed by using a linear prediction parameter representing a spectrum envelope of the speech signal and an excitation parameter corresponding to the linear prediction residual signal. There is a known method. Since the speech coding method using such a linear prediction method can obtain relatively high quality synthesized speech with a small transmission capacity, various application methods are actively studied in conjunction with recent advances in hardware technology. , Has been developed.

[0003] Among such speech coding methods, a method that can obtain good sound quality is described in, for example, "Improved speech qualit by Kleijin and the like.
yand efficient vector qua
ntization inSELP "(ICASSP '
88 s4.4, pp. 155-158, 1988), CELP (Code E).
xcited Linear Predictive
Coding) method is well known.

[0004] In recent years, variable bit rate coding has been proposed which performs more efficient coding in non-voice sections and unvoiced sections. S. P5,41
No. 4,796. Such variable bit rate coding not only has the advantage that the total bit rate can be reduced, but also transmits or records information about the state of audio as coded data. This has the advantage that the state of the audio can be directly known and the possibility of application at the time of decoding is expanded.

[0005] It is known that a speech decoding apparatus for the above-described speech encoding apparatus uses a post-filter for the purpose of improving subjective sound quality. This post filter utilizes the auditory masking characteristics,
This is to perform noise shaping. Noise shaping refers to a function of transforming quantization noise, which originally has almost flat spectral characteristics, so as to be close to the spectral characteristics of the speech to be processed, and suppressing the perception of the quantization noise by masking. Such a post-filter is generally provided immediately after the speech decoder.

FIG. 10 shows a configuration in which a post filter is provided immediately after a CELP decoding device.

[0007] In FIG.
Is connected to a first input terminal of an adder 115 via a multiplier 112, and the probability codebook 113
4 is connected to the second input terminal of the adder 115. The output terminal of the adder 115 is connected to the post filter 101 via the synthesis filter 116 and to the adaptive codebook 111 via the delay circuit 117. Further, the demultiplexer 118 is connected to the adaptive codebook 111, the probability codebook 113, the synthesis filter 116, and the post filter 101, respectively.

In the configuration described above, the demultiplexer 118 decomposes the received signal into a linear prediction parameter a, a delay L and a gain b of an adaptive codebook, and an index i and a gain g of a probability codebook. The parameter a is applied to the synthesis filter 116 and the post filter 101, the delay L is applied to the adaptive codebook 111 and the post filter 101, the gain b is applied to the multiplier 112,
The index i and the gain g are respectively assigned to the probability codebook 11
3 and output to the multiplier 114.

The code vector of the adaptive codebook 111 is selected based on the delay L of the adaptive codebook output from the demultiplexer 118. Here, the adaptive codebook 111 has the same content as the content of the adaptive codebook in the encoding device. A past drive sound source signal is input to adaptive codebook 111 via delay circuit 117. The multiplier 112 amplifies the input adaptive code vector with the gain b and sends the amplified adaptive code vector to the adder 115.

On the other hand, a code vector of the probability codebook 113 is selected based on the index i of the probability codebook output from the demultiplexer 118.
Here, the probability codebook 113 has the same content as the content of the probability codebook in the encoding device. Multiplier 11
4 amplifies the input code vector by the gain g and sends it to the adder 115.

The adder 115 creates a drive excitation signal by adding the amplified probability code vector and the amplified adaptive code vector, and sends it to the synthesis filter 116 and the delay circuit 117.

The synthesis filter 116 performs a synthesis process on the driving excitation signal using the received linear prediction parameter a as a coefficient, and outputs a decoded signal.

The post-filter 101 performs spectrum shaping on the input decoded signal using the linear prediction parameter a and the delay L of the adaptive codebook.

The post filter 101 has a known structure, for example, Juin-Hway Chen, Al.
"Adaptive Po" by len Gersho
stfiltering for Quality E
nhancement of Coded Spec
h "(IEEE Transactions of Sp)
ech and Audio Processin
g, Vol. 3, No. 1, pp. 59-71, Jan
uary, 1995).

For example, according to this example, the post filter 101 comprises a pitch emphasis filter 121, a formant emphasis filter 122, a spectrum tilt correction filter 123, and a gain control section 124, as shown in FIG. As shown in (1), they are connected in cascade.

The formant emphasis filter 122
Emphasizes the first, second, and third formants X, Y, and Z of the spectral envelope shown in FIG. 5 derived from the speech waveform shown in FIG. Vowels are characterized by the formant peak values X, Y, and Z. The dotted line in FIG. 5 indicates the slope of the spectrum.

Here, the transfer function Hp (z) of the pitch emphasis filter 121, the transfer function Hs (z) of the formant emphasis filter 122, and the spectrum tilt correction filter 123
Are represented by the following equations, respectively.

[0018]

(Equation 1) Here, p is a pitch period (the number of samples), and γ and λ are weighting coefficients.

[0019]

(Equation 2) Here, a1 is a spectrum parameter, n is a linear prediction order, α and β are weighting coefficients, and usually 0 <β <
It is set in the range of α <1.

[0020]

(Equation 3) Here, μ is a correction coefficient and is set in a range of 0 <μ <1.

The pitch component of the reproduced sound is emphasized by the pitch emphasizing filter having the transfer function of the above equation 1, and has a role of increasing clarity. In addition, the formant enhancement filter having the transfer function of Equation 2 enhances the formant structure of the reproduced voice, and the energy of the quantization noise concentrates on this formant portion, and becomes close to the spectrum envelope of the reproduced voice. It has the role of making it difficult to perceive the formation noise. The spectrum tilt correction filter having the transfer function of Equation 3 has a role of removing the tilt of the spectrum that could not be corrected by the formant emphasis filter.

[0022]

However, the above-described post-filter is intended only to improve subjective sound quality for voiced sounds having a pitch structure and a formant peak. Consonants and non-speech sections without pitch structure and formant peaks,
In particular, when post-filtering is performed in a non-speech section in a situation where background noise is mixed, an original pitch or formant is emphasized by a pitch emphasis filter and a formant emphasis filter.

Background noise represented by air conditioners and automobiles generally has a slope spectrum such that it has large energy in a low frequency band and energy gradually decreases in a high frequency band. For such background noise, the high band is rather emphasized by the spectrum tilt correction filter. Due to these, there is a problem that the sound quality of the reproduced sound becomes very uncomfortable.

The present invention has been made in view of such a problem, and an object of the present invention is to provide a post filter capable of obtaining a reproduced sound without a sense of incongruity even when an input signal is non-voice.

[0025]

To achieve the above object, a first post-filter of the present invention comprises a pitch structure control means for controlling a pitch structure of an input signal based on a pitch parameter; Formant structure control means for controlling a formant structure based on speech / non-speech information and a spectrum parameter for distinguishing whether an input signal is speech or non-speech, and a slope of a spectrum of an output signal of the formant structure control means. Spectrum tilt correcting means for correcting, gain control means for controlling the gain of the output signal of the formant structure control means,
Selecting means for executing the pitch structure control means based on the non-speech information when the input signal is speech, and controlling not to execute the pitch structure control means when the input signal is non-speech.

In order to achieve the above object, the second aspect of the present invention
The post-filter of the first post-filter further enhances the formant when the input signal is a speech based on the speech / non-speech information, and flattens the formant when the input signal is non-speech. As described above, a first coefficient control means for adaptively changing the formant control coefficient in the formant structure control means is provided.

In order to achieve the above object, the third aspect of the present invention
The post-filter of the first or second post-filter further corrects the slope of the spectrum when the input signal is speech based on the speech / non-speech information, and the input signal is non-speech. In some cases, a second coefficient control means for adaptively changing the correction coefficient of the spectrum tilt in the spectrum tilt correction means so as to maintain the spectrum tilt is provided.

The first post-filter includes a pitch structure control means for controlling a pitch structure of the input signal based on a pitch parameter, and a formant structure control means for controlling whether the input signal is a voice signal or a non-voice signal. Control the formant structure based on the speech / non-speech information and the spectral parameters that distinguish between The spectrum tilt correcting means corrects the tilt of the spectrum of the output signal of the formant structure control means, and the gain control means controls the gain of the output signal of the formant structure control means. When the input signal is voice based on the voice / non-voice information, the pitch structure control means is executed, and when the input signal is non-voice, the pitch structure control means is controlled not to execute the pitch structure control means. .

The second post-filter has, in addition to the function of the first post-filter, a first coefficient control means which, when the input signal is a voice based on the voice / non-voice information, forms a formant. When the input signal is non-voice, the formant control coefficient in the formant structure control means is adaptively changed so as to flatten the formant.

In the third post filter, in addition to the operation of the first or second post filter, an input signal is a voice based on the voice / non-voice information by a second coefficient control means. Sometimes, the spectrum tilt is corrected, and when the input signal is non-voice, the spectrum tilt correction means adaptively changes the spectrum tilt correction coefficient so as to maintain the spectrum tilt.

[0031]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing a configuration of a post filter according to an embodiment to which the present invention is applied.
Note that the post filter of the present embodiment is described as being provided immediately after the variable bit rate encoding / decoding device. Also, the variable bit rate encoding / decoding device determines whether the audio state of the input signal in the frame to be processed is speech or non-speech, and encodes the signal at a predetermined bit rate depending on whether it is speech or non-speech. Device to be encrypted / decoded.

FIG. 2 is a block diagram showing a configuration of a variable bit rate CELP decoding apparatus to which the post-filter of the above-described embodiment shown in FIG. 1 is applied. 5 shows a configuration provided with a post filter.

First, the configuration of the variable bit rate CELP decoding device will be briefly described. As shown in the figure,
The adaptive codebook 11 is connected to the adder 1 via the multiplier 12.
5 is connected to the first input terminal of
Is connected to the second input terminal of the adder 15 via the multiplier 14. The output terminal of the adder 15 is connected to the post filter 1 shown in FIG. 1 via the synthesis filter 16 and to the adaptive codebook 11 via the delay circuit 17. Note that post filter 1
Will be described later in detail.

On the other hand, the variable bit rate CELP decoding device includes a demultiplexer 18, and the demultiplexer 18 includes the adaptive codebook 11, the probability codebook 13, the synthesis filter 16, and the postfilter 1.
Are connected to each other.

Further, the demultiplexer 18 includes:
The voice state determination unit 19 is connected.

The demultiplexer 18 converts the coded data signal received via the voice state determination unit 19 into a linear prediction parameter a which is a spectrum parameter, a delay L which is the same as a delay L of an adaptive codebook which is a pitch parameter, and a probability It is decomposed into a codebook index i and the same gain g. Then, the decomposed linear prediction parameter a is applied to the synthesis filter 16 and the postfilter 1, the delay L of the adaptive codebook is applied to the adaptive codebook 11 and the postfilter 1, and the gain b is applied to the multiplier 12.
Then, the index i and the gain g of the probability codebook are output to the probability codebook 13 and the multiplier 14, respectively.

On the other hand, the apparatus selects a code vector of the adaptive codebook 11 based on the delay L of the adaptive codebook output from the demultiplexer 18. Here, the adaptive codebook 11 has the same content as the content of the adaptive codebook in the encoding device.

On the other hand, a past driving sound source signal is input to the adaptive codebook 11 via a delay circuit 17. Further, the multiplier 12 receives a gain b from the demultiplexer 18, amplifies the adaptive code vector input from the adaptive codebook 11 based on the gain b, and sends out the amplified code vector to the adder 15.

The code vector of the probability codebook 13 is selected based on the index i of the probability codebook output from the demultiplexer 18. Here, the probability codebook 13 has the same content as the content of the probability codebook in the encoding device. Further, the gain g is input to the multiplier 14 from the demultiplexer 18, and the probability code vector input from the probability code book 13 is amplified based on the gain g,
The data is sent to the adder 15.

The adder 15 creates a drive excitation signal by adding the amplified probability code vector and the amplified adaptive code vector, and sends it to the synthesis filter 16 and the delay circuit 17. .

The synthesis filter 16 performs a synthesis process on the driving excitation signal using the received linear prediction parameter a as a coefficient, and outputs a decoded signal.

The post filter 1 performs spectrum shaping on the input decoded signal using the linear prediction parameter a and the delay L of the adaptive codebook.

Also, in FIG.
Reads out information related to the voice state, that is, voice / non-voice information f from the coded data, reads out a predetermined amount of coded data based on the information f, and outputs the voice / non-voice information f to the post-filter 1. It is supposed to. The demultiplexer 18 decomposes the audio / non-speech information f into parameters with a predetermined bit allocation according to the speech / non-speech information f, and then performs a decoding process.

Returning to FIG. 1, a post filter according to a first embodiment to which the present invention is applied will be described.

The post filter 1 of this embodiment includes a selection switch 2 as a selection means, a pitch control filter 3 as a pitch structure control means for controlling a pitch structure, a formant control filter 4 as a formant structure control means, The apparatus includes a spectrum tilt correction filter 5 as a spectrum tilt correction unit, a gain control unit 6 as a gain control unit, and a coefficient control unit 7 as first and second coefficient control units.

In the post filter of this embodiment, the transfer function Hp1 (z) of the pitch control filter 3, the transfer function Hs1 (z) of the formant control filter 4, and the transfer function Ht1 (z) of the spectrum tilt correction filter 5 are respectively It is represented by the following equation.

[0048]

(Equation 4) Here, p is a pitch period (the number of samples), and γ and λ are weighting coefficients.

[0049]

(Equation 5) Here, ai is a linear prediction parameter which is a spectrum parameter, n is a linear prediction order, and α and β are weighting coefficients. In the present embodiment, control is performed in a range of 0 <β <α <1.

[0050]

(Equation 6) Here, μ is a correction coefficient, and in this embodiment, 0 <μ
It is controlled within the range of <1.

As described above, the pitch control filter 3 has a transfer function represented by the above equation (4). Then, information indicating whether the received sound is sound or non-speech, and whether or not to execute the sound is selected by the selection switch 2 that operates based on the sound / non-speech information f. When the sound received from the voice / non-voice information f is “voice”, the selection switch 2 controls the pitch control filter 3.
Is executed, and when the pitch control filter 3 is executed, the pitch component of the reproduced sound is emphasized and the clarity is increased.

The formant control filter 4 has a transfer function represented by the above equation (5). The coefficient is controlled by a coefficient control unit 7 that operates based on the voice / non-voice information f. This coefficient control unit 7
Indicates that the received sound is “voice” from the voice / non-voice information f.
Is determined, the coefficients α and β of the formant control filter 4 are controlled to emphasize the formant. Thereby, the formant structure of the reproduced sound is emphasized, the spectrum of the quantization noise is controlled, and the quantization noise is reduced.

On the other hand, when the coefficient control unit 7 determines that the received sound is “non-voice” from the voice / non-voice information f, the coefficient control unit 7 changes the coefficients α and β of the formant control filter 4 to flatten the formant. Control.

Further, the spectrum tilt correction filter 5 has a transfer function represented by the above equation (6). The coefficient is controlled by a coefficient control unit 7 that operates based on the voice / non-voice information f. When the coefficient control unit 7 determines that the received sound is “voice” from the voice / non-voice information f, the coefficient control unit 7 controls the coefficient μ of the spectrum tilt correction filter 5 to correct the tilt of the spectrum. Thereby, the inclination of the spectrum that cannot be corrected by the formant control filter 4 is corrected.

On the other hand, when the coefficient control unit 7 determines that the received sound is “non-voice” based on the voice / non-voice information f, the coefficient μ of the spectrum tilt correction filter 5 maintains the tilt of the spectrum. Is appropriately controlled.

Hereinafter, the operation of the post filter having the above-described configuration according to the present embodiment will be specifically described with reference to a flowchart shown in FIG.

The post-filter 1 includes the synthesizing filter 16
, And the voice / non-voice information f from the voice state determination unit 19, the value of the voice / non-voice information f is checked, and the decoded signal from the synthesis filter 16 is converted into voice. It is determined whether there is a voice or non-voice (step S1). In the present embodiment, voice / non-voice information f = 1
If f = 0, it is determined that the voice is non-voice.

In step S1, the determination is yes.
If so, the switch 2 is closed to the terminal b side, and the pitch control filter 3 is executed (step S2). Next, the coefficient control unit 7 sets the weighting coefficients α and β of the formant control filter 4 to 0.8 and 0.5, respectively (step S3), and sets the coefficient μ of the spectrum tilt correction filter 5 to 0.
Set to 5 (step S4).

Thereafter, the formant control filter 4 (step S7), the spectral tilt correction filter 5 (step S8), and the gain control unit 6 (step S9) are operated, and output as post-filtered signals (step S10). .

The determination in step S1 is n.
If it is o, that is, if it is non-voice, the pitch control filter 3 is not executed by closing the switch 2 to the terminal c side. Next, the coefficient control unit 7 sets the superposition coefficients α and β of the formant control filter 4 to 0.5 and 0.1, respectively, and sets the correction coefficient μ of the spectrum tilt correction filter 5 to 0.1 in step S5 (step S6). . Thereafter, as described above, the formant control filter 4 (step S7), the spectrum tilt correction filter 5
(Step S8), the gain control unit 6 (Step S9) is executed, and output as a post-filtered signal (Step S10).

In general, "speech" has a relatively large difference between the strength of speech and the strength of quantization noise depending on the frequency band as shown in FIG. There is a band where the difference from the noise intensity is small and the sound is hard to hear. When such a speech is input, the post-filter of the present embodiment brings the spectral characteristics of the quantization noise closer to the spectral characteristics of the speech as shown in FIG.

Further, when "speech" is input, the slope of the spectrum characteristic is corrected by the spectrum slope correction filter 5 so that the slope of the spectrum characteristic of the quantization noise does not deviate from the slope of the spectrum characteristic of the speech. I do.

On the other hand, when "non-speech", for example, background noise of an automobile or air conditioner as shown in FIG. 8 is input, the signal passes through a conventional spectrum tilt correction filter and becomes "A" in FIG.
As shown by, the spectral characteristics tend to be emphasized in a high frequency band. In consideration of such a point, the post filter of the present embodiment performs control so as to maintain the slope of the spectrum so that a high frequency band is not emphasized when “non-voice” is input as described above. ing. Thereby, the characteristics are improved as indicated by "B" in FIG.

As described above, according to the post filter of the present embodiment, when “voice” is input by the above-described processing, pitch enhancement is performed by the pitch control filter 3 as in the related art, and the formant The structure is emphasized, and the spectrum tilt correction filter 5 corrects the tilt of the spectrum.
When “non-voice” is input, the pitch control filter 3
Is performed, and the formant structure is flattened by the formant control filter 4 so that the spectrum of the spectrum inclination correction filter 5 is controlled to maintain a gentle inclination.

As described in detail above, according to the post-filter of the above embodiment, it is possible to obtain a reproduced sound that does not cause a sense of incongruity even in a consonant part or a non-voice section having no pitch structure or formant peak.

[0066]

As described above, according to the first aspect of the present invention, it is possible to provide a post filter capable of obtaining a reproduced sound without a sense of incongruity even when an input signal is non-voice.

Further, according to the second aspect of the present invention, it is possible to provide a post filter capable of obtaining a reproduced sound that does not cause a sense of incongruity even in a consonant part or a non-voice section having no pitch structure and formant peak.

Further, according to the third aspect of the present invention,
In a consonant part or a non-speech section having no pitch structure and formant peak, a post-filter capable of obtaining a reproduced sound without a sense of incongruity can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a post filter according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a variable bit rate CELP decoding apparatus to which the post filter of the embodiment shown in FIG. 1 is applied.

FIG. 3 is a flowchart showing a specific operation of the post filter of the embodiment.

FIG. 4 is a diagram showing an example of an audio waveform.

FIG. 5 is a diagram showing an example of a spectral envelope.

FIG. 6 is a diagram illustrating an example of an input “voice” characteristic in the post filter of the embodiment.

FIG. 7 is a diagram showing spectral characteristics after processing of quantization noise when the “voice” shown in FIG. 6 is input to the post filter of the embodiment.

FIG. 8 is a diagram showing an example of an input “non-speech” characteristic in the post filter of the embodiment.

FIG. 9 is a diagram illustrating the spectral characteristics of a reproduced signal when non-voice is input in the post filter of the embodiment and the conventional post filter.

FIG. 10 is a block diagram showing a configuration example of a conventional post filter.

FIG. 11 is a block diagram showing a configuration of a variable bit rate CELP decoding apparatus to which the post filter shown in FIG. 10 is applied.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 post filter 2 selection switch 3 pitch control filter 4 formant control filter 5 spectral tilt correction filter 6 gain control unit 7 coefficient control unit 11 adaptive codebook 13 probability codebook 16 synthesis filter 17 Delay circuit 18 Demultiplexer 19 Voice state determination unit

Claims (3)

[Claims] 1. A pitch structure control means for controlling a pitch structure of an input signal based on a pitch parameter, voice / non-voice information and spectrum for distinguishing whether the input signal is voice or non-voice. Formant structure control means for controlling the formant structure based on the parameters; spectrum inclination correction means for correcting the slope of the spectrum of the output signal of the formant structure control means; and gain for controlling the gain of the output signal of the formant structure control means Control means for executing the pitch structure control means based on the speech / non-speech information when the input signal is speech, and selecting not to execute the pitch structure control means when the input signal is non-speech; Means, and a post filter. 2. A formant control unit for enhancing formants when an input signal is voice based on the voice / non-voice information and flattening the formants when the input signal is non-voice. The post filter according to claim 1, further comprising: first coefficient control means for adaptively changing coefficients. 3. The spectrum tilt correcting means for correcting a slope of a spectrum when the input signal is a voice based on the voice / non-voice information, and for maintaining a tilt of the spectrum when the input signal is a non-voice. 3. The post filter according to claim 1, further comprising: a second coefficient control unit that adaptively changes a correction coefficient of a spectrum tilt in the above.