JP3350340B2 - Voice coding method and voice decoding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio encoding method for compressing and encoding an audio signal and an audio decoding method for decoding an original audio signal from encoded data.

[0002]

2. Description of the Related Art A technology for efficiently compressing and encoding a voice signal at a low bit rate is an important technology for effective use of radio waves and reduction of communication costs in mobile communications such as car telephones and in-company communications. It is. A CELP (Code Excited Linear Prediction) method is known as a speech coding method capable of performing high-quality speech synthesis at a bit rate of 8 kbps or less. The CELP method is described in "Code-Excit" by Mr. MR Schroeder and BSAtal of AT & T Bell Labs.
ed Linear Prediction (CELP) High-Quality Speech at
Very low Bit Rates ”, Proc.ICASSP; 1985, pp.937-939”
Since it was announced in (Reference 1), it has attracted attention as a method capable of synthesizing high-quality speech, and various studies have been made on improving quality, reducing the amount of calculation, and the like.

An adaptive codebook is one of the components of a speech coding apparatus based on the CELP system. The adaptive codebook is
The pitch prediction analysis of the input voice is performed by closed loop operation or analysis by synthesis (Analysis by Synthesis). Generally, pitch prediction analysis using an adaptive codebook requires 20
The pitch period is searched in a search range of 147 samples (128 candidates) to find a pitch period that minimizes distortion with respect to the target signal, and the information of the pitch period is often transmitted as 7-bit encoded data. .

However, if the input speech signal contains a pitch cycle that is out of the above search range, the pitch cycle cannot be represented by the adaptive codebook. There is a problem that deterioration occurs. Further, if the pitch period search range of the adaptive codebook is to be expanded to prevent this, it is necessary to increase the number of bits of the encoded data representing the pitch period, resulting in a problem that the transmission rate increases. there were.

[0005]

As described above, in the conventional speech coding method, the pitch period of a predetermined search range is coded into coded data having a predetermined number of bits. If a voice having a pitch period out of the range is input, quality degradation occurs. In general, the range of the pitch period to be coded is experimentally verified and selected appropriately, but there is no guarantee that the range will always fall within this range, and the characteristics of the speaker and the fluctuation of the pitch period within the same speaker In some cases, the danger of being out of the search range of the pitch period is always present.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems of the prior art, and a speech encoding method and speech decoding capable of correctly expressing a pitch period of a speech signal to obtain high quality speech. The purpose of the present invention is to provide a conversion method.

[0007]

In order to solve the above problems, a speech encoding method according to the present invention analyzes a pitch cycle of an input speech signal and suppresses a pitch cycle component of the input speech signal. When the pitch period of the input audio signal is searched for and encoded based on the audio signal passed through the perceptual weight filter including the pitch filter, a range of the pitch period that can be expressed by the encoded data is applied to the pitch filter. The analysis range of the given pitch period is set wide. More specifically, the range of the pitch cycle (TL) that can be represented by the encoded data of the pitch cycle is TLL ≦ TL ≦ TLH, and the analysis range of the pitch cycle (TW) given to the pitch filter is TWL ≦ TW ≦ TW
When H, at least one of the conditions of TLL> TWL and TLH <TWH is satisfied.

The perceptual weight filter improves the subjective quality by making the quantization noise inaudible using the masking effect. Here, the masking effect refers to a phenomenon that in a frequency region where the power spectrum of the input voice is large, even if the quantization noise is large, it is masked by the spectrum of the input voice and becomes hard to hear. Conversely, in a frequency region where the power spectrum of the input signal is small, the masking effect does not work, and quantization noise is easily heard. The audibility weighting filter has a function of shaping the spectrum of the quantization noise so as to be close to the spectrum of the input voice, and an LPC synthesis filter corresponding to the spectrum envelope of the voice, and an input voice signal corresponding to the fine structure of the voice spectrum. And a pitch filter having a function of suppressing the pitch period component of

Conventionally, an analysis range (TWL ≦ TW) of a pitch period (TW) given to a pitch filter in an auditory weighting filter has been described.
W ≦ TWH) is the pitch period (TL) according to the adaptive codebook
(TLL ≦ TL ≦ TLH).
That is, TWL = TLL and TWH = TLH. This is based on the premise that a valid search range of the pitch cycle by the adaptive codebook has already been selected, and the analysis range of the pitch cycle given to the pitch filter is made to match the pitch cycle search range of the adaptive codebook. There is nothing other than that.

[0010] By the way, since the perceptual weighting filter is used as a distortion measure for codebook search in the voice coding device, it is not necessary to send information representing the configuration of the perceptual weighting filter to the voice decoding device. Therefore, the analysis range of the pitch period given to the pitch filter in the perceptual weight filter can be set freely, unlike the pitch period search range of the adaptive codebook limited by the number of bits of the encoded data. Focusing on this point, in the present invention, the analysis range of the pitch cycle given to the pitch filter in the perceptual weighting filter is set sufficiently wider than the pitch cycle search range of the adaptive codebook.

In this way, even if an input speech signal having a pitch period that cannot be represented by the pitch period search range of the adaptive codebook is provided, the pitch period to be applied to the pitch filter can be accurately determined. Therefore, based on this pitch cycle, the pitch filter suppresses the pitch cycle component of the input speech signal, and the auditory weighting filter including the pitch filter performs the spectrum shaping of the quantization noise, thereby improving the speech quality due to the masking effect. Can be planned. In addition, since this process does not change the connectivity between the speech encoding device and the speech decoding device, quality can be improved while maintaining compatibility.

On the other hand, a speech decoding method according to the present invention analyzes a pitch cycle of a decoded speech signal obtained by decoding encoded data, and converts the decoded speech signal into a post signal including a pitch filter for emphasizing a pitch cycle component. A speech decoding method for outputting through a filter is characterized in that an analysis range of a pitch cycle given to a pitch filter is set wider than a range of a pitch cycle that can be represented by encoded data. More specifically, the range of the pitch period (TL) that can be represented by the encoded data is TLL ≦ TL ≦ TLH,
When the analysis range of the pitch cycle (TP) given to the pitch filter is TPL ≦ TP ≦ TPH, at least one of TLL> TPL and TLH <TPH is satisfied.

The post-filter is a filter that enhances the subjective quality by emphasizing peaks and attenuating valleys of the spectrum of the decoded speech signal obtained by the speech decoding device. As one component of this post filter,
There is a pitch filter that emphasizes the pitch period component of the decoded audio signal.

Conventionally, as information on a pitch period (TP) used for a pitch filter in this post filter, a pitch period obtained by decoding encoded data is used as it is, or a pitch period search range of an adaptive codebook is used. Either a new pitch period is obtained as the same range as above. In any case, the analysis range of the pitch period (TP) ｛TPL ≦ TP ≦
TPH} is the same range as the search range {TLL ≦ TL ≦ TLH} of the pitch period (TL) of the adaptive codebook. That is, TPL = TLL and TPH = TLH. This is based on the premise that a valid search range for the pitch period by the adaptive codebook has already been selected.
This is because the analysis range of the pitch cycle given to the pitch filter matches the pitch cycle search range of the adaptive codebook.

Since the post-filter performs processing on the decoded speech signal, the analysis range of the pitch period given to the pitch filter in the post-filter is an adaptive codebook limited by the number of bits of the encoded data. Unlike the pitch period search range, the pitch period can be set freely. Focusing on this point, in the present invention, the analysis range of the pitch cycle applied to the pitch filter in the post filter is set to be sufficiently wider than the range of the pitch frequency that can be represented by the encoded data, that is, the pitch cycle search range of the adaptive codebook.

In this way, even if a decoded speech signal having a pitch period that cannot be represented by the pitch period search range of the adaptive codebook is given to the post-filter, if the pitch period of the decoded speech signal is obtained, the The pitch cycle component is emphasized based on the pitch cycle, and the pitch cycle component that could not be transmitted can be restored to improve the quality.

[0017]

DESCRIPTION OF THE PREFERRED EMBODIMENTS First, an embodiment of a speech encoding method according to the present invention will be described. FIG. 1 is a block diagram for explaining a basic operation of an audibility weighting filter used in a speech encoding method according to an embodiment of the present invention. In FIG. 1, a digitized audio signal (input audio signal) is sequentially input from an input terminal 101 in frame units with a plurality of samples as one frame. In this embodiment, one frame has 80 samples. This input audio signal is L
It is provided to the PC coefficient analysis unit 102, the pitch information analysis unit 103, and the audibility weight filter 104.

The LPC coefficient analysis unit 102 analyzes the input speech signal on a frame-by-frame basis using an existing technique such as the autocorrelation method, and calculates the LPC coefficient ｛α (i); i = 1 to N
Find P｝. In this LPC analysis, it is necessary to use data having a length sufficient to obtain a stable analysis result centering on the analysis target frame of the input audio signal. NP
Represents the analysis order, and NP = 10 here. LPC coefficient {α (i) thus obtained; i = 1 to NP}
Is the pitch information analysis unit 103 and the perceptual weight filter 1
04.

The pitch information analysis unit 103 analyzes the input speech signal on a frame basis, for example, as described later, and obtains a pitch period TW and a pitch filter coefficient g. The details of the pitch information analysis unit 103 will be described later with reference to FIG.

The audibility weighting filter 104 is a filter for shaping the spectrum of the quantization noise so as to be close to the spectrum of the input voice signal, and corresponds to an LPC synthesis filter corresponding to the voice spectrum envelope and a voice spectrum fine structure. And a pitch filter having a function of suppressing a pitch cycle component of the input audio signal. Specifically, the audibility weighting filter 104 is the LPC coefficient analysis unit 1
LPC coefficient ｛α (i) obtained from Eq. 02
Based on P｝ and pitch cycle TW and pitch filter coefficient g obtained from pitch information analysis section 103, (1)
A filter having a transfer function W (z) defined by the equation is configured to perform filtering on an input audio signal input in frame units, and output a weighted input audio signal to an output terminal 105.

[0021]

(Equation 1)

A (z / β) / A (z / γ) corresponds to an audibility weighting filter corresponding to the spectral envelope of speech, and
(Z) corresponds to an audibility weighting filter corresponding to the spectral fine structure of the voice. Here, as specific numerical values of each parameter, the present inventors have β = 0.9, γ = 0.4, λ
= 0.4 is recommended. However, since the values of these parameters depend on subjective preferences, these values are not always optimal. A weighted input audio signal obtained by passing the input audio signal through an audibility weighting filter 104 having a transfer function W (z) defined by equation (1) is output from an output terminal 105.
Output.

Next, referring to FIG.
03 will be described. In FIG. 2, an input terminal 301
From the input audio signal, from the input terminal 302 LP
The C coefficient {α (i); i = 1 to NP} is input and provided to the prediction residual signal calculation unit 303. The prediction residual signal calculation unit 303 performs analysis using data having a length sufficient to obtain a stable analysis result centering on the analysis target frame of the input speech signal, similarly to the LPC coefficient analysis unit 102. The prediction residual signal calculation unit 303 converts the data of the input speech signal used for analysis into ｛u (n); n = 0 to NU−
1}, LPC coefficient {α (i); i = 1 to NP}
And the prediction residual signal ｛e for the data u (n)
(N); n = 0 to NU-1} is obtained according to the following equation.

[0024]

(Equation 2)

The prediction residual signal {e (n); n = 0 to NU-1} obtained in this manner is supplied to the pitch period analyzer 304. The pitch cycle analysis unit 304 calculates a signal {ew (n) obtained by multiplying the prediction residual signal {e (n); n = 0 to NU-1} by a Hamming window, n = 0 to NU-1}. The autocorrelation value φ (t) defined by the equation (6) is calculated in the pitch period analysis range {TWL ≦ t ≦ TWH}.

[0026]

(Equation 3)

In this embodiment, the lower limit TWL and the upper limit TWH of the pitch period analysis range are set to, for example, TWL = 10
And TWH = 200. On the other hand, a pitch period search range {TLL ≦ TL ≦ TL of a pitch period encoding means (for example, an adaptive codebook described later) not shown in FIG.
The lower limit TLL and the upper limit TLH of H｝ are, for example, TLL =
20, TLH = 147. That is, TLL> TW
L, TLH <TWH, and the characteristic is that the pitch cycle analysis range is wider than the pitch cycle search range.

The autocorrelation value φ (t) thus obtained
Is given to the pitch filter coefficient analysis unit 305 as the pitch period TW. In pitch filter coefficient analysis section 305, prediction residual signal {e (n); n = 0 to NU-1} obtained in prediction residual signal calculation section 303.
And pitch cycle TW obtained by pitch cycle analysis section 304
Is used to calculate the pitch filter coefficient g according to the following equation.

[0029]

(Equation 4)

The pitch period TW and the pitch filter coefficient g thus obtained are output from the output terminal 306. Here, the case where the first-order pitch filter is used has been described, but it may be realized by using a higher-order pitch filter. In this case, although the amount of calculation slightly increases, more accurate pitch information can be obtained. Further, the pitch period analysis method and the pitch filter coefficient analysis method are not limited to the methods described above, and other techniques may be used.

The above processing is summarized as a flowchart shown in FIG. First, step S
At 11, the LPC coefficient {α (i); i = 1 to NP} is obtained,
Next, in step S12, the prediction residual signal ｛e (n); n = 0
~ NU-1}. In step S13, the pitch period T
W is analyzed, and then a pitch filter coefficient g in the pitch cycle TW is obtained in step S14. In step S15, steps S11, S13 and S
14, the LPC coefficient {α (i); i = 1 to NP},
Using the pitch period TW and the pitch filter coefficient g, a perceptual weight filter defined by the expression (1) is formed, and in step S16, the input voice signal is passed through the perceptual weight filter to generate and output a weighted input voice signal. .

Next, a description will be given of a speech coding apparatus based on the CELP system using the above-described perceptual weighting filter with reference to FIG. In FIG. 4, the same components as those in FIG. 1 are denoted by the same reference numerals, and detailed description will be omitted.

First, LPC coefficients {α (i); i = 1 to NP} output from LPC coefficient analysis section 102 are provided to LPC coefficient quantization section 106 and quantized. In the weighted synthesis filter 107, information of the LPC coefficient {α (i); i = 1 to NP} from the LPC coefficient analysis unit 102, information of the pitch period TW and pitch filter coefficient g from the pitch information analysis unit 103, LPC coefficient quantum By receiving the information of the quantized LPC coefficient {αq (i); i = 1 to NP} from the conversion unit 106, a filter having a transfer function of Hw (z) is configured. This weighted synthesis filter 107
Is represented by the following equation.

Hw (z) = W (z) · H (z) (8) Here, the transfer function W (z) of the audibility weighting filter 104
Is the same as equation (1) shown above. The synthesis filter H (z) is expressed by the following equation.

[0035]

(Equation 5)

The drive signal applied to the weighted synthesis filter 107 is represented by a combination of each of the adaptive codebook 108, adaptive vector gain codebook 113, noise codebook 109, and noise vector gain codebook 111.

The adaptive codebook 108 always holds the immediately preceding drive signal sequence, generates an adaptive vector by repeating this drive signal sequence at a desired pitch cycle, and efficiently expresses the periodicity. However, since this pitch period must be transmitted via the multiplexer, the pitch period is searched only within the range of the number of candidates that can be represented by a predetermined number of bits. In the present embodiment, the adaptive codebook 108
, The pitch period search range {TLL ≦ TL ≦ TLH} is set as TLL = 20 and TLH = 147.

The noise codebook 109 is a codebook having a noise sequence as a candidate vector. Generally, the random codebook 10
9 is structured in order to reduce the amount of calculation and improve the quality, but is not particularly related to the gist of the present invention, so that the description is omitted.

An adaptive vector and an adaptive vector gain are selected from adaptive codebook 108 and adaptive vector gain codebook 113, respectively, and are multiplied by multiplier 110. Similarly, an adaptive vector and a noise vector gain are selected from the noise codebook 109 and the noise vector gain codebook 114, respectively, and are multiplied by the multiplier 111. Then, the adders 112, 1
A drive signal is generated by adding the vectors of the 11 outputs to each other, and is provided to the weighted synthesis filter 107 as an input.

Using the output signal of the perceptual weighting filter 104 as a target signal, the error of the output signal of the weighted synthesis filter 107 is obtained by a subtractor 115, and further, the minimum distortion search unit 11
6, the square distortion is calculated. The minimum distortion search unit 116 calculates an adaptive vector, an adaptive vector gain, and a noise vector that minimize the square distortion with respect to the adaptive codebook 108, the adaptive vector gain codebook 113, the noise codebook 109, and the noise vector gain codebook 111. And a combination of the noise vector gain and the adaptive vector when the square distortion is minimized, the adaptive vector gain, the noise vector, and index information of each candidate of the noise vector gain are provided to the multiplexer 117.

On the other hand, the LPC coefficient quantization section 106
Index information obtained as a result of quantizing the coefficients is supplied to the multiplexer 117. Multiplexer 117
Are the LPC coefficient quantization unit 106 and the minimum distortion search unit 11
6 is converted into a bit stream as encoded data and output to an output terminal 118. Finally, a drive signal when the square distortion calculated by the minimum distortion search unit 116 is minimized is given to the adaptive codebook 108 to update the internal state, and prepare for the input speech signal of the next frame.

As described above, in the present embodiment, the pitch period analysis range ΔT of the pitch information analysis unit 103 used in the audibility weighting filter 104 and the weighted synthesis filter 107.
WL ≦ TW ≦ TWH} and the weighted synthesis filter 107
Represents the periodicity of the drive signal applied to the
Of the pitch period search range {TLL ≦ TL ≦ TLH} of the adaptive codebook 108 represented by encoded data of the pitch period (encoded data of the adaptive vector index) which is encoded via the output terminal 7 and output from the output terminal 118 Relationship is TWL
<TLL and TWH> The condition of TLH is satisfied.
That is, pitch cycle analysis range ｛TWL ≦ TW ≦ TW
H} is set wider than the pitch period search range {TLL ≦ TL ≦ TLH}.

By satisfying such a condition, the pitch period search range of the adaptive codebook 108 which must be represented by a predetermined number of bits {TLL ≦ TL ≦ T
Even if an input speech signal having a pitch cycle deviating from LH is given, the analysis range of the pitch filter in the audibility weighting filter 104 and the weighted synthesis filter 107 ｛TW
Since L ≦ TW ≦ TWH} is wider than the pitch period search range of the adaptive codebook 108, the spectrum of the quantization noise can be shaped at the pitch period of the input speech signal, and the noise effect can be reduced by the masking effect. As a result, the subjective quality can be effectively improved.

In this embodiment, the pitch period analysis range {TWL ≦ TW ≦ TWH} is changed to the pitch period search range {TL
The relationship with L ≦ TL ≦ TLH} is TLL> TWL and T
Although the condition of LH <TWH is satisfied, TLL> TWL
Alternatively, only one of the following conditions may be satisfied: TLH <TWH.

Next, an embodiment of a speech decoding method according to the present invention will be described. FIG. 5 is a block diagram for explaining a basic operation of the post filter used in the speech decoding method according to one embodiment of the present invention. In the figure, a digitized audio signal (for example, a decoded audio signal) is sequentially input from an input terminal 201 in frame units with a plurality of samples as one frame. In this embodiment, one frame is described as 80 samples.

On the other hand, from the input terminal 202, the input terminal 2
The LPC prediction residual signal of the audio signal from S01 or a signal similar thereto, for example, a drive signal for driving a synthesis filter in a CELP-type audio decoding device described later. The pitch information analysis unit 203 obtains a pitch cycle using the LPC prediction residual signal or the driving signal of the synthesis filter. Details of the pitch information analysis unit 203 will be described later.

The post filter 205 has an input terminal 20
For example, a decoded speech signal is given from 1; information on a pitch period TP and a pitch filter coefficient g is given from a pitch information analyzer 203; and information on LPC coefficients {α (i); i = 1 to NP} from an input terminal 204. Is given. LP
The C coefficient represents the spectrum envelope of the audio signal from the input terminal 201. The post filter 205 determines the pitch period TP and the LPC coefficient ｛α (i); i = 1 to N
A filter represented by the following transfer function R (z) is configured using the information of P｝, and the audio signal from the input terminal 201 is subjected to filtering processing. The filtered output signal is output from output terminal 206. R (z) = F (z) · P (z) · U (z) (10) F (z), P (z) and B (z) are expressed as follows.

[0048]

(Equation 6)

Here, as specific numerical values of each parameter, the present inventors have ν = 0.5, ξ = 0.8, η = 0.
7, μ = 0.4 is recommended. The values of these parameters are
These numbers are not always optimal, as they depend on subjective preferences.

Next, the pitch information analyzer 203 in the present embodiment will be described with reference to FIG. 6, the same components as those in FIG. 2 are denoted by the same reference numerals, and description thereof will be omitted.

The pitch information analyzer 203 shown in FIG.
The difference between the first embodiment and the pitch information analyzer 103 shown in FIG. 2 in the previous embodiment lies in the input signal. That is, the prediction residual signal or a signal equivalent thereto, for example, a drive signal generated by a speech decoding device (not shown) is input to the pitch information analysis unit 203 in FIG. Therefore, it is not necessary to input the input speech signal and the LPC coefficient to the pitch information analysis unit 203 as in the case of the pitch information analysis unit 103 in FIG.
No three is needed. Pitch information analyzer 203 in FIG.
Is the pitch period T obtained by the pitch period analysis unit 304
Information about P and the pitch filter coefficient g obtained by the pitch filter coefficient analysis unit 305 is output from an output terminal 308.

Here, the lower limit TPL and the upper limit TP of the analysis range {TPL ≦ TP ≦ TPH} of the pitch cycle TP in the pitch cycle analysis section 304 in the pitch information analysis section 203.
H is set to, for example, TPL = 10 and TPH = 200. On the other hand, the lower limit TLL and the upper limit TLH of the pitch period search range {TLL ≦ TL ≦ TLH} of the pitch period encoding means (for example, an adaptive codebook) are TLL = 20, TLH
= 147. That is, TLL> TPL, TPH>
TLH, which is characterized in that the pitch cycle analysis range is wider than the pitch cycle search range.

The above processing is summarized as a flowchart shown in FIG. First, step S
At 21 the pitch cycle TP is analyzed, and then at step S22 a pitch filter coefficient g at the pitch cycle TP is determined. In step S23, steps S21 and S
A post filter defined by equation (10) is configured using the pitch period TP and pitch filter coefficient g obtained in step 22 and the LPC coefficient input from the input terminal 204,
In step S24, the audio signal input from the input terminal 201 is output through a post filter.

Next, a speech decoding apparatus based on the CELP system using the above-described post filter will be described with reference to FIG. 8, the same components as those in FIG. 5 are denoted by the same reference numerals, and detailed description will be omitted.

In FIG. 8, a bit stream which is coded data output from a speech coding apparatus based on a CELP system (not shown) is input to an input terminal 401 via a transmission path or a storage medium (not shown). The speech encoding device is configured, for example, as shown in FIG. The demultiplexer 402 decodes parameters necessary for generating an audio signal from the input bit stream.
The type and number of these parameters vary depending on the configuration of the speech coding apparatus. Here, the LPC coefficient index, the adaptive vector index, the adaptive vector gain index, the noise vector index, and the noise vector gain index are used as parameters. It shall be decrypted.

An adaptive vector and an adaptive vector gain specified by an adaptive vector index and an adaptive vector gain index are selected from the adaptive codebook 403 and the adaptive vector gain codebook 404, and are multiplied by a multiplier 405. Similarly, a noise vector and a noise vector gain specified by the noise vector index and the noise vector gain index are selected from the noise codebook 406 and the noise vector gain codebook 407, and these are selected by the multiplier 40.
Multiplied by eight.

Then, the adder 409 causes the multiplier 40
The drive signal is generated by summing the vectors output from 5,408, and the drive signal is provided to the synthesis filter 411 and the pitch information analyzer 203. This drive signal is also given to the adaptive codebook 403 to update the internal state and prepare for the next input.

On the other hand, when the LPC coefficient index is given to LPC coefficient decoding section 410, LPC coefficient {αq (i); i = 1 to NP} is decoded.
The coefficients are calculated by the synthesis filter 411 and the post filter 205.
Given to. The transfer function of the synthesis filter 411 is (9)
This is expressed in the same manner as in the expression, and filtering is performed by inputting a drive signal from the adder 409 to the synthesis filter 411, so that a decoded speech signal is obtained. This decoded audio signal is input to the post filter 205.

The post filter 205 and the pitch information analyzer 203 are as described with reference to FIGS. 5 to 7, and a detailed description thereof will be omitted. In the present embodiment, the decoded voice signal output from the synthesis filter 411 is input to the post filter 205, and the drive signal output from the adder 409 is input to the pitch information analysis unit 203. In the audio decoding device of the present embodiment, the post filter 205
The decoded audio signal passed through is finally output from the output terminal 206.

As described above, in the present embodiment, the pitch period analysis range ΔTPL ≦ T in the pitch information analysis unit 203 for analyzing the pitch information used in the post filter 205
P ≦ TPH}, which represents the periodicity of the drive signal supplied to the synthesis filter 411, and can be a pitch period (TL) specified by the adaptive vector index used by the adaptive codebook 403 after being decoded via the demultiplexer 402. When the relationship of the range {TLL ≦ TL ≦ TLH} is TPL <T
LL and TPH> TLH are satisfied. That is, the pitch cycle analysis range {TPL ≦ TP ≦ TPH} is set wider than the pitch cycle range {TLL ≦ TL ≦ TLH} that can be expressed by encoded data of the pitch cycle (encoded data of the adaptive vector index). .

By satisfying such a condition, the pitch period search range of the adaptive codebook 403 which must be represented by a predetermined number of bits {TLL ≦ TL ≦ T
Even if a decoded speech signal having a pitch period deviating from LH is input to the post filter 205, the pitch analysis range {TPL ≦ TP ≦ TPH} of the pitch information analysis unit 203 used in the post filter 205 is determined by the adaptive codebook 40.
Since the pitch period is wider than the pitch period search range of 3, the pitch period that cannot be transmitted as encoded data of the adaptive vector index can be restored. As a result, the subjective quality can be improved.

In the present embodiment, the relationship with the pitch cycle range {TLL ≦ TL ≦ TLH} in which the pitch cycle analysis range {TPL ≦ TP ≦ TPH} can be represented by encoded data is TPL <TLL and TPH> TLH. Although the condition is satisfied, only one of TLL> TPL and TLH <TPH may be satisfied.

Next, another embodiment of the present invention will be described.
FIG. 9 is a block diagram for explaining a basic operation of a post filter used in a speech encoding method according to another embodiment of the present invention. In the figure, the same components as those of FIG. 5 are denoted by the same reference numerals, and detailed description thereof will be omitted.

The difference between this embodiment and the embodiment shown in FIG. 5 is that a speech decoding apparatus (not shown) has an adaptive codebook and a fixed codebook made up of fixed candidate vectors prepared in the same column, and The point that the method of obtaining the pitch period TP is different between the case where the adaptive codebook is selected and the case where the fixed codebook is selected.

When the adaptive codebook is selected, the pitch period TL of the transmitted and decoded adaptive codebook is regarded as a pitch period TP to be applied to the pitch filter in the post filter.
Using this pitch period TP, a pitch filter coefficient g is obtained and given to the post filter 205. Conversely, when the fixed codebook is selected, the pitch information analysis unit 203 newly obtains a pitch cycle TP, calculates a pitch filter coefficient g using the pitch cycle TP, and provides the calculated pitch filter coefficient g to the post filter 205.

Next, the pitch information analyzer 203 in the present embodiment will be described with reference to FIG. 10, the same components as those in FIG. 6 are denoted by the same reference numerals, and description thereof will be omitted.

In FIG. 10, selection information indicating which of an adaptive codebook and a fixed codebook is used in a speech decoding device (not shown) is provided from an input terminal 211. When this selection information indicates an adaptive codebook, switch 3
Reference numeral 10 denotes a pitch filter coefficient analysis unit 3 which regards information on the pitch period TL of the adaptive codebook input from the input terminal 207 as information on the pitch period TP used in the post filter.
Give to 05. On the other hand, when the selection information from the input terminal 211 indicates a fixed codebook, the switch 310
It works to make the input from 2 valid. That is, a drive signal sequence that is a prediction residual signal or a signal equivalent thereto is input from the input terminal 202, a pitch period TP is obtained by the pitch period analysis unit 304 based on this signal, and is provided to the pitch filter coefficient analysis unit 305. . At this time, the fixed codebook is selected because it can be considered that a pitch that cannot be expressed by the pitch period search range {TLL ≦ TL ≦ TLH} of the adaptive codebook has occurred. The analysis range is ｛TPL ≦ TP <TLL, TLH <TP ≦ T excluding the pitch period search range of the adaptive codebook.
PH｝, and as a result, the amount of calculation required for analyzing the pitch period can be reduced.

The pitch filter coefficient analysis unit 305 calculates a pitch filter coefficient g using a prediction residual signal or a drive signal sequence equivalent thereto based on the information of the given pitch cycle TP, and calculates the pitch cycle TP and the pitch filter coefficient. The information of g is output from the output terminal 308.

The above processing is summarized as a flowchart shown in FIG. Step S3 in FIG.
3, S34, S35, and S36 are the same as steps S21, S22, S23, and S24 in FIG. 7, and thus description thereof is omitted here. However, step S33
Is different from the pitch cycle analysis range in step S21 as described above.

First, at step S31, it is determined whether the selection information indicates an adaptive codebook or a fixed codebook. When the selection information indicates the adaptive codebook, the process proceeds to step S32, and when the selection information indicates the fixed codebook, the process proceeds to step S33. If the selection information indicates an adaptive codebook, the pitch period TL obtained in the adaptive codebook search in step S32
Is the pitch period TP used in the pitch filter in the post filter, and the process proceeds to step S34. If the selection information indicates a fixed codebook, the pitch cycle TP is newly calculated in step S33, and the process proceeds to step S34.

Next, a speech decoding apparatus based on the CELP method using the above-described post filter will be described with reference to FIG. 12, the same components as those in FIG. 8 are denoted by the same reference numerals, and detailed description will be omitted.

This embodiment is different from the embodiment of FIG. 8 in that a fixed codebook 412 is provided in the same column as adaptive codebook 403. The following description focuses on the differences from the embodiment of FIG.

In FIG. 12, the demultiplexer 402
The adaptive vector index output from
3 and the vector to be decoded is the adaptive codebook 4
03 or a fixed codebook 412
Is determined to be a vector generated from. The determination result is given to the switch 414, the switch 415, and the pitch information analysis unit 203. Here, the adaptive vector index is the adaptive codebook 403 and the fixed codebook 412.
Although the case where both vectors are expressed in the same way has been described, the determination information may be directly generated from the demultiplexer, and in that case, the determination unit 413 is unnecessary. In this case, a speech coding apparatus (not shown) has a configuration in which decision information is newly given to the multiplexer as information to be transmitted, and the decision information includes one bit to indicate the distinction between the adaptive codebook and the fixed codebook. Additional information is required.

Switch 414 switches between providing an adaptive vector index to adaptive codebook 403 and fixed codebook 412 based on the determination information provided from determination section 413. Similarly, the switch 415 is connected to the determination unit 41
The vector to be given to the multiplier 405 is determined based on the judgment information given from 3.

In the pitch information analysis section 203, the judgment section 41
3, the calculation method of the pitch period TP of the pitch filter used in the post filter 205 is switched as described with reference to FIGS.
The pitch period TP and the pitch filter coefficient g obtained by the pitch information analysis unit 203 are given to the post filter 205.

Next, effects of the present embodiment will be described. Adaptive codebook 403 generates an adaptive vector that can efficiently represent the pitch period using the immediately preceding drive signal sequence, whereas fixed codebook 412 prepares a plurality of predetermined fixed vectors. The pitch period of the input speech signal given to the speech coding device (not shown) is the pitch period search range of adaptive codebook 403 {TLL ≦ TL ≦ TLH}.
, The adaptive vector of the adaptive codebook 403 is selected and its index is encoded.

However, if the input speech signal is
, The fixed codebook 412 is used instead of the adaptive codebook 403. This is whether adaptive codebook 403 was used,
This means that it is possible to determine whether or not the pitch period of the input speech signal is included in the pitch period search range of the adaptive codebook 403 depending on whether the fixed codebook 412 is used.

Further, when the fixed codebook 412 is used, the pitch cycle analysis range of the pitch information analysis unit 203 is the pitch cycle search range of the adaptive codebook 403 ｛TLL ≦ TL
≦ TLH}, the pitch period analysis range is set to {TPL ≦ TP <TLL, TLH <TP ≦
TPH}, and the amount of calculation can be reduced. On the other hand, when the adaptive codebook 403 is selected, it is considered that the pitch cycle of the input speech signal is represented by the pitch cycle TL of the adaptive codebook 403, and the pitch filter in the post filter 205 based on the pitch cycle TL. You only need to emphasize the pitch.

Although the above-described embodiment has described the example in which the present invention is applied to the CELP-based speech encoding / decoding method, the present invention relates to an APC (Adaptive Predictive Codin).
The present invention can also be applied to other types of speech encoding / decoding methods such as the g) method.

[0080]

As described above, according to the present invention, it is possible to provide a speech encoding method and a speech decoding method capable of correctly expressing the pitch period of a speech signal and obtaining high-quality speech.

That is, in the speech coding method according to the present invention, the analysis range of the pitch period given to the pitch filter in the perceptual weight filter is set wider than the pitch period search range of the adaptive codebook, so that the pitch period of the adaptive codebook is Even if an input voice signal having a pitch period that cannot be represented by the search range is provided, the pitch period to be applied to the pitch filter can be accurately obtained. Therefore, based on this pitch period, the pitch filter suppresses the pitch period component of the input audio signal, and the spectral weighting of the quantization noise is performed by the auditory weighting filter including the pitch filter.
The sound quality can be improved by the masking effect. In addition, since this process does not change the connectivity between the speech encoding device and the speech decoding device, quality can be improved while maintaining compatibility.

In the speech decoding method according to the present invention,
By setting the analysis range of the pitch period given to the pitch filter in the post filter wider than the range of the pitch frequency that can be represented by the encoded data, the decoded speech signal having the pitch period that cannot be represented by the encoded data is sent to the post filter. Even if given, if the pitch period of the decoded speech signal is obtained, the pitch period component is emphasized based on the pitch period, and the pitch period component that could not be transmitted can be restored to improve the quality. .

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining a basic operation of an auditory weighting filter used in a speech encoding method according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a pitch information analyzer in the embodiment.

FIG. 3 is a flowchart showing a processing procedure according to the embodiment;

FIG. 4 is a diagram illustrating C to which the speech encoding method according to the embodiment is applied;
FIG. 2 is a block diagram showing a configuration of a speech synthesis device using the ELP method

FIG. 5 is a block diagram for explaining a basic operation of a post filter used in a speech decoding method according to an embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration of a pitch information analyzer in the embodiment.

FIG. 7 is a flowchart showing a processing procedure according to the embodiment;

FIG. 8 shows C to which the speech decoding method according to the embodiment is applied.
FIG. 2 is a block diagram showing a configuration of a speech decoding device based on the ELP method.

FIG. 9 is a block diagram illustrating a basic operation of a post-filter used in the speech decoding method according to the embodiment;

FIG. 10 is a block diagram showing a configuration of a pitch information analysis unit in the embodiment.

FIG. 11 is a flowchart showing a processing procedure according to the embodiment;

FIG. 12 is a block diagram showing a configuration of a CELP-based speech decoding apparatus to which a speech decoding method according to another embodiment of the present invention is applied.

[Description of Codes] 101: Audio signal input terminal 102: LPC coefficient analysis unit 103: Pitch information analysis unit 104: Perceptual weight filter 105: Output terminal 106: LPC coefficient quantization unit 107: Weighted synthesis filter 108: Adaptive codebook 109: noise codebook 110, 111: multiplier 112: adder 113: adaptive vector gain codebook 114: noise vector gain codebook 115: subtractor 116: minimum distortion search unit 117: multiplexer 118: output terminal 201: decoded voice Signal input terminal 202 ... Drive signal input terminal 203 ... Pitch information analysis unit 204 ... LPC coefficient input terminal 205 ... Post filter 206 ... Post filter output terminal 207 ... Adaptive codebook pitch period input terminal 211 ... Codebook selection information input terminal 301 ... Sound signal input terminal 302 ... LPC Number input terminal 303: prediction residual signal calculation unit 304: pitch cycle analysis unit 305: pitch filter coefficient input terminal 306 ... output terminal 308 ... output terminal 310 ... switch 401 ... bit stream input terminal 402 ... demultiplexer 403 ... adaptive codebook 404 ... Adaptive vector gain codebook 405 ... Multiplier 406 ... Noise codebook 407 ... Noise vector gain codebook 408 ... Multiplier 409 ... Adder 410 ... LPC coefficient decoding unit 411 ... Synthesis filter 412 ... Fixed codebook 413 ... Decision unit 414, 415 ... switch

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-9-81192 (JP, A) JP-A-6-202698 (JP, A) JP-A-8-76793 (JP, A) 77834 (JP, A) (58) Field surveyed (Int. Cl. ⁷ , DB name) G10L 19/04 G10L 19/12

Claims (4)

(57) [Claims] An input audio signal is analyzed for a pitch period and is applied to a pitch filter for suppressing a pitch period component of the input audio signal. Based on the audio signal passed through an auditory weighting filter including the pitch filter, the input signal is analyzed. In a speech encoding method for searching and encoding a pitch cycle of an audio signal, a range of a pitch cycle that can be represented by encoded data of the pitch cycle is set to be wider than a range of an analysis of a pitch cycle given to the pitch filter. Characteristic speech coding method. 2. A pitch filter for analyzing a pitch cycle of an input voice signal and applying the same to a pitch filter for suppressing a pitch cycle component of the input voice signal, based on the voice signal passed through an auditory weighting filter including the pitch filter. In a speech encoding method for searching for and encoding a pitch cycle of a speech signal, a pitch cycle (TL) that can be represented by the encoded data
Is defined as TLL ≦ TL ≦ TLH, and the analysis range of the pitch cycle (TW) given to the pitch filter is defined as TWL ≦ T
When W ≦ TWH, TLL> TWL and TLH <
A speech coding method characterized by satisfying at least one condition of TWH. 3. Speech decoding for analyzing a pitch cycle of a decoded speech signal obtained by decoding encoded data and outputting the decoded speech signal through a post filter including a pitch filter for emphasizing a pitch cycle component. A speech decoding method, wherein an analysis range of a pitch cycle given to the pitch filter is set wider than a range of a pitch cycle that can be represented by the encoded data. 4. Speech decoding for analyzing a pitch cycle of a decoded speech signal obtained by decoding encoded data and outputting the decoded speech signal through a post filter including a pitch filter for emphasizing a pitch cycle component. A pitch period (TL) that can be represented by the encoded data.
Is defined as TLL ≦ TL ≦ TLH, and the analysis range of the pitch period (TP) given to the pitch filter is TPL ≦ T
When P ≦ TPH, TLL> TPL and TLH <
A speech decoding method characterized by satisfying at least one condition of TPH.