JP2000250597A - Lsp correcting device, voice encoding device, and voice decoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain auditory equal effects over the entire frequency band by correcting an LSP(line spectrum pair) converted into the frequency range corresponding to auditory characteristics and putting the frequency range of the LSP back to a linear frequency range. SOLUTION: A bark conversion part 1 when inputting the LSP converts the respective dimensional values of the LSP from the linear frequency range into a bark frequency range and outputs a bark-LSP as the conversion result to an LSP deformation part 2. An inter-dimension distance calculation part 3 of the LSP deformation part 2 once receiving the bark-LSP from the bark conversion part 1 calculates inter-adjacent-dimension distances of the bark-LSP in order and outputs the inter-adjacent-dimension distances to an inter-dimension distance expansion part 4 of the LSP conversion part 2. The expansion part 4 once receiving the inter-adjacent-dimension distance of the bark-LSP from the inter-dimension distance calculation part 3 compares it with a specific threshold, and corrects a dimension value when the threshold is not exceeded and performs a process for widening the inter-adjacent-dimension distance, thereby outputting the corrected bark-LSP to a bark reconversion part 5.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an LSP correction device for correcting an LSP (line spectrum pair), which is a spectrum parameter used by a voice encoding device and a voice decoding device, and to compress a digital voice signal into a small amount of information. The present invention relates to an audio encoding device and an audio decoding device that decodes an audio code to reproduce a digital audio signal.

[0002]

2. Description of the Related Art Conventional speech coding apparatuses and speech decoding apparatuses using an LSP as a spectrum parameter are disclosed in JP-A-4-5700 and JP-A-5-2739.
No. 97 is disclosed. JP-A-4-5
Japanese Patent Application Publication No. 700 (hereinafter referred to as Conventional Example 1) has the same structure in a speech encoding device and a speech decoding device for the purpose of suppressing quality degradation of decoded sound due to LSP encoding errors and transmission path errors. An LSP correction means is provided.

As the LSP correction method, a method is disclosed in which a distance between adjacent dimensions is calculated, and when the distance is smaller than a threshold, the interval is increased to the threshold.
Specifically, the LSP to be corrected is ω (k), k = 1 to M
Then, the distance d (k) between adjacent dimensions is as shown below. d (k) = ω (k + 1) −ω (k)

When the distance d (k) between adjacent dimensions falls below the threshold value D, an LSP correction process is executed as described below. Here, ω ′ is the corrected LSP. ω ′ (k) = {ω (k) + ω (k + 1)} / 2−D / 2 ω ′ (k + 1) = {ω (k) + ω (k + 1)} / 2 + D / 2

Japanese Unexamined Patent Publication No. Hei 5-273997 (hereinafter referred to as Conventional Example 2) discloses an instability at the time of analysis which occurs when a decoded sound having deteriorated quality (quantization noise) is re-analyzed to calculate a spectrum parameter. For the purpose of reducing the influence of quantization and quantization noise, in the backward type CELP-based speech coding apparatus and speech decoding apparatus, the spectral parameters obtained by the linear prediction analysis means are converted on the LSP. LSP control means for performing the above control is provided.

[0006] As the LSP control means, the calculated LS
A method of linearly adding P and an LSP fixedly given in advance using a coupling coefficient is disclosed. In particular,
Coupling coefficient beta, if the fixed LSP and omega _0, LSP obtained by linear addition becomes as follows. ω ′ (k) = ω (k) · β + ω ₀ (k) (1−β)

Further, the second conventional example discloses that an allowable limit value D (k) of the distance between adjacent dimensions of the LSP used for controlling the value of the coupling coefficient β is given for each dimension. The setting of the permissible limit value D (k) is performed based on the distance between adjacent dimensions of the LSP obtained from a linear prediction coefficient obtained by previously analyzing an input speech that does not include quantization noise. In Conventional Example 2, the fixed value D (k) is compared with the distance d (k) between adjacent dimensions, and d (k)> D for all orders.
The coupling coefficient β is controlled so as to satisfy (k). d (k) = ω (k + 1) −ω (k)

A conventional post-filter (post-processing filter in an audio decoding device) using the LSP correction process is disclosed in Japanese Patent Application Laid-Open No. 8-305397. Japanese Patent Application Laid-Open No. 8-305397 (hereinafter referred to as Conventional Example 3) discloses a correction LS calculated by an LSP correction process in order to increase the degree of freedom in designing a post filter.
The voice emphasis processing is performed by P.

As the LSP correction process, there are disclosed an internal division process on the LSP represented by the same formula as that of the conventional example 2 and a process of increasing the distance between adjacent dimensions similar to the conventional example 1. However, here, when the distance between adjacent dimensions is smaller than the threshold value, the distance between adjacent dimensions is increased to the threshold value by shifting the higher-order LSP collectively above that portion, and all the adjacent dimensions (in order from the lower order) As a result of performing the processing for, the method of uniformly reducing the distance between all adjacent dimensions by the total distance shifted upward is disclosed.

[0010]

Since the conventional speech coding apparatus and speech decoding apparatus are constructed as described above, in the case of the prior art examples 1 and 3, the correction of the LSP for increasing the distance between adjacent dimensions is performed. When executing the process, a fixed threshold is used regardless of the value of the input LSP. However, in the low frequency band and the high frequency band, even though the bandwidth is the same in the linear frequency domain, the bandwidth of the basilar membrane inside the auditory organ is greatly different, and the correction performed with a fixed threshold in the linear frequency domain The effect is very different between low and high frequencies. For this reason, there has been a problem that appropriate correction is not performed over the entire band.

For example, consider a case where an LSP correction process is introduced to reduce the effect of LSP quantization error in a speech encoding device. Even if the amplitude instability can be suppressed in the low frequency range by using the threshold value D, there may be a case where the same threshold value D cannot suppress the steep pole to such an extent as to be auditoryly noticeable in the high frequency range. Conversely, if the threshold D is set large enough to suppress the steep poles in the high band, the poles in the low band will be too wide and the average coding distortion will be greatly degraded. As described above, in the conventional inter-dimension distance extension processing using a threshold independent of the frequency band, deterioration of the coding / decoding quality of the applied speech coding apparatus and speech decoding apparatus cannot be eliminated, or conversely, There was a problem of causing deterioration.

[0012] Further, consider a case where connections are made in multiple stages, such as a speech coding apparatus, a speech decoding apparatus, a speech coding apparatus, and a speech decoding apparatus. In this case, the first speech coding apparatus and the first speech decoding apparatus introduce a certain degree of deterioration or spectrum deformation due to quantization noise, post-filter, or the like. For this reason, in the second speech encoding device and the speech decoding device, an LSP having a steep pole that does not exist in the normal input speech may be generated, and unless this is appropriately corrected. An encoded / decoded sound with a large sense of distortion is generated. A similar problem occurs when background noise is superimposed. As described above, the conventional LSP correction means has a problem in that even when there is a steep pole that causes such a sense of distortion in a high frequency band, it cannot be sufficiently corrected as described above. To repeat, if the LSP correction is set strong enough to suppress the steep pole in the high band, there is a problem that the pole in the low band is too wide and the average coding distortion is greatly deteriorated.

When the LSP correction processing is executed using the internal division processing disclosed in Conventional Examples 2 and 3, it is possible to adaptively control the coupling coefficient β or set a value for each dimension. Also, since the correction is performed irrespective of the value on the frequency of the LSP to be corrected, the perceptual effect of this correction differs for each frequency, and the coding of the applied voice coding apparatus and voice decoding apparatus is performed. There has been a problem that the deterioration of the decoding quality cannot be eliminated, or on the contrary, the deterioration occurs.

In the second conventional example, the allowable limit value D (k) of the distance between adjacent dimensions is given in advance for each dimension and used for controlling the coupling coefficient β. , One permissible limit value D per dimension
Setting (k) increases the error.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and has as its object to provide an LSP correction device capable of providing an acoustically uniform effect over the entire frequency band. Another object of the present invention is to provide a speech coding apparatus capable of improving coding quality. Another object of the present invention is to provide a speech decoding device capable of improving the decoding quality.

[0016]

An LSP correction apparatus according to the present invention corrects an LSP converted to a frequency domain corresponding to auditory characteristics, and returns the LSP frequency domain to a linear frequency domain. It is.

[0017] The LSP correction apparatus according to the present invention comprises an LSP
Is converted from the linear frequency domain to the Bark frequency domain to make correction.

The LSP correction device according to the present invention is an LSP correction device.
Is converted from the linear frequency domain to the mel frequency domain for correction.

An LSP correction device according to the present invention is an LSP correction device.
Is converted from the linear frequency domain to the logarithmic frequency domain for correction.

An LSP correction device according to the present invention is an LSP correction device.
The threshold value for the distance between adjacent dimensions is calculated for each dimension on the basis of each dimension value of LSP, and the LSP is calculated based on the threshold value for each dimension.
Is corrected.

An LSP correction apparatus according to the present invention
Is converted from the linear frequency domain to the frequency domain corresponding to the auditory characteristics, and the threshold value is calculated in the frequency domain corresponding to the auditory characteristics.

An LSP correction apparatus according to the present invention compares a threshold calculated in a frequency domain corresponding to an auditory characteristic with a threshold defined in a linear frequency domain, and outputs a larger threshold to the correction means. It was done.

The LSP correction device according to the present invention is an LSP correction device.
Is converted from the linear frequency domain to the Bark frequency domain to calculate the threshold.

The LSP correction device according to the present invention is an LSP correction device.
Is converted from the linear frequency domain to the mel frequency domain to calculate the threshold value.

The LSP correction device according to the present invention is an LSP correction device.
Is converted from the linear frequency domain to the logarithmic frequency domain to calculate the threshold.

A speech encoding apparatus according to the present invention encodes an LSP corrected in a frequency domain corresponding to auditory characteristics and outputs an LSP code and a quantized LSP.

[0027] The speech encoding apparatus according to the present invention comprises an LSP
The corrected LSP is encoded using a threshold value for each dimension calculated based on each dimension value of
The SP is output.

[0028] A speech encoding apparatus according to the present invention calculates an encoded sound source from a decoded LSP corrected in a frequency domain corresponding to auditory characteristics and an input speech.

[0029] The speech encoding apparatus according to the present invention comprises an LSP
The encoded excitation is calculated from the decoded LSP and the input speech corrected using the threshold value for each dimension calculated on the basis of each dimension value.

The speech decoding apparatus according to the present invention is configured to generate a synthesized sound from a decoded LSP corrected in a frequency domain corresponding to the auditory characteristics and a sound source signal generated by a sound source decoding unit.

[0031] The speech decoding apparatus according to the present invention comprises an LSP.
The synthesized sound is generated from the decoded LSP corrected by using the threshold value for each dimension calculated based on each dimension value and the sound source signal generated by the sound source decoding means.

The speech decoding apparatus according to the present invention executes a spectrum emphasis process on a synthesized sound using a decoded LSP corrected in a frequency domain corresponding to auditory characteristics.

[0033] The speech decoding apparatus according to the present invention comprises an LSP
The spectrum emphasizing process is performed on the synthesized sound using the decoded LSP corrected using the threshold value for each dimension calculated based on each dimension value.

[0034]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a block diagram showing an LSP correction apparatus according to Embodiment 1 of the present invention. In FIG. 1, when an LSP which is a spectrum parameter is input, the frequency domain of the LSP is shifted from the linear frequency domain to ba.
a bark transformation unit (conversion unit) 2 for converting to an rk frequency region (a frequency region corresponding to auditory characteristics); an LSP transformation unit (correction unit) 3 for correcting the LSP whose frequency domain has been transformed by the bark transformation unit 1; Is an inter-dimensional distance calculating unit that calculates the distance between adjacent dimensions of the LSP, and 4 is an inter-dimensional distance that increases the distance between adjacent dimensions when the distance between adjacent dimensions calculated by the inter-dimensional distance calculating unit 3 falls below a predetermined threshold. The extension unit 5 is a bark inverse transform unit (inverse transform unit) that returns the frequency domain of the LSP corrected by the LSP transforming unit 2 to the linear frequency domain.

Next, the operation will be described. First, LSP
Is input to the bark transform unit 1, and the LSP is obtained by being calculated or decoded by a speech coding device or a speech decoding device, and is used to stabilize LSP parameters and to flatten a spectrum. Is input to the LSP correction device for the purpose of conversion.

When the LSP is input, the bark transform unit 1 converts each dimension value of the LSP from the linear frequency domain into a bar.
k-frequency domain, and the result of the conversion, bark-
The LSP is output to the LSP transformation unit 2. The conversion from the linear frequency domain to the bark frequency domain includes a method of calculating based on a table lookup and a method of performing conversion according to an approximate expression as shown below.

[0037]

(Equation 1)

When the distance-to-dimension calculating unit 3 of the LSP transforming unit 2 receives the bark-LSP from the barrier transforming unit 1, it calculates the distance between adjacent dimensions of the bark-LSP in order, and calculates the distance between adjacent dimensions. Is output to the inter-dimensional distance extension unit 4 of the LSP transformation unit 2.

Upon receiving the distance between adjacent dimensions of the bark-LSP from the interdimensional distance calculation section 3, the interdimensional distance extension section 4 compares the distance between adjacent dimensions with a predetermined threshold value D.
If the distance between adjacent dimensions is less than the threshold D, bar
A process for correcting the corresponding dimension value of the k-LSP to increase the distance between adjacent dimensions is performed, and the corrected bark-LS
P is output to the inverse inverse transform unit 5.

Specifically, the bark-LSP to be corrected is
Where b (k) and k = 1 to M, the distance d between adjacent dimensions
(K) is as shown below, and when the distance d (k) between adjacent dimensions falls below the threshold value D, as shown below, ba
The rk-LSP correction process is performed. Here, b ′ is the corrected bark-LSP. d (k) = b (k + 1) -b (k) b '(k) = {b (k) + b (k + 1)} / 2-D / 2b' (k + 1) = {b (k) + b (k + 1) )｝ / 2 + D / 2

The extension processing in the inter-dimension distance extension unit 4 is not limited to this, but is described in Japanese Patent Laid-Open No. 8-305.
Various methods such as the method disclosed in Japanese Patent No. 397 can be used. Further, the configuration of the LSP deforming unit 2 is not limited to the configuration of the first embodiment.

When receiving the corrected bark-LSP from the LSP transforming unit 2, the bark inverse transform unit 5 performs an inverse transform process for returning each dimension value of the corrected bark-LSP from the bark frequency domain to the linear frequency domain. And outputs a corrected LSP as a result of the inverse conversion. In addition, ba
The conversion from the rk frequency domain to the linear frequency domain includes a method of calculating based on a table lookup and a method of performing conversion according to an approximate expression as shown below. This expression corresponds to the inverse conversion of the approximate expression in the bark conversion unit 1.

[0043]

(Equation 2)

FIG. 2 is an explanatory diagram for explaining a correction result of the LSP correction device according to the first embodiment. FIG. 2A shows an input LSP. The input LSP has a sixth order for the sake of simplicity, but generally an LSP of the order of 8 to 14 is used. FIG. 2B is an example of a bark-LSP obtained by converting the input LSP by the bark converting unit 1. ba
When converted to the rk frequency domain, the distance between adjacent dimensions in the low frequency band increases, and the distance between adjacent dimensions in the high frequency band decreases. This corresponds well to the human auditory sensation. FIG.
(C) is an example of a corrected bark-LSP obtained as a result of performing deformation of the bark-LSP by the LSP deforming unit 2. The inter-dimensional distance between ω5 and ω6 is widened based on the threshold value D regarding the distance between adjacent dimensions in the bark frequency domain.

When the same LSP deformation processing is performed in the linear frequency domain as in the conventional case, the distance between ω1 and ω2 and the distance between ω2 and ω3 in FIG. However, this is aurally ω5 and ω6
The pole that exists between the two is felt most steeply, and the response is poor. According to the method of the first embodiment, the interval between ω5 and ω6 can be favorably increased.

As is clear from the above, the first embodiment
According to the method described above, since the correction process is performed after converting the frequency domain of the input LSP from the linear frequency domain to the bark frequency domain, an effect that an auditory equal effect can be obtained over the entire frequency band. Play. Further, even when the speech encoding device and the speech decoding device are connected in multiple stages or when there is a lot of background noise, it is possible to satisfactorily correct a steep pole generated in a high frequency band without causing deterioration in a low frequency region. It works.

When the LSP correction device using the approximation formula of the bark conversion unit 1 is applied to a speech coding device or a speech decoding device, for example, there is a tendency that the degradation of the high-frequency coding characteristic is large. On the contrary, when the low-frequency correction is weak and the amplitude of the decoded sound becomes unstable, the linear frequency domain and b
The approximation formula can be adjusted so as to convert to an intermediate frequency domain of the ark frequency domain. In this case, the bark inverse transform unit 5 also needs to be adjusted accordingly.

Embodiment 2 FIG. 3 is a block diagram showing an LSP correction apparatus according to Embodiment 2 of the present invention. In FIG. 3, when an LSP which is a spectrum parameter is input, the frequency domain of the LSP is changed from a linear frequency domain to a mel frequency domain (audio Transform unit (conversion means) for converting the LSP into a frequency domain corresponding to the dynamic characteristic, 7 is an LSP transformation unit (correction means) for correcting the LSP whose frequency domain has been transformed by the mel transform unit 6, and 8 is an adjacent dimension of the LSP. When the distance between adjacent dimensions calculated by the distance calculating unit 8 is smaller than a predetermined threshold, the distance between dimension calculating unit 9 for calculating the distance between dimensions is expanded by the dimension distance expanding unit 9 for expanding the distance between adjacent dimensions. L
This is a mel inverse transform unit (inverse transform unit) that returns the frequency domain of the LSP corrected by the SP transforming unit 7 to the linear frequency domain.

Next, the operation will be described. First, LSP
Is input to the mel transform unit 6, where the LSP is obtained by being calculated or decoded by a speech coding device or a speech decoding device, etc., and is used to stabilize the LSP parameters and flatten the spectrum. Is input to the LSP correction device for the purpose of conversion.

Upon receiving the LSP, the mel transform unit 6 converts each dimension value of the LSP from the linear frequency domain to the mel frequency domain, and outputs the mel LSP, which is the result of the conversion, to the LSP transforming unit 7. The conversion from the linear frequency domain to the mel frequency domain includes a method of calculating based on a table lookup and a method of performing conversion according to an approximate expression as shown below.

[0051]

(Equation 3)

Upon receiving the mel LSP from the mel conversion unit 6, the inter-dimensional distance calculation unit 8 of the LSP transformation unit 7 calculates the distance between adjacent dimensions of the mel LSP in order, and calculates the distance between adjacent dimensions to the LSP. Output to the inter-dimension distance extension unit 9 of the deformation unit 7.

Upon receiving the distance between adjacent dimensions of the mel LSP from the distance calculating unit 8, the distance expanding unit 9 compares the distance between adjacent dimensions with a predetermined threshold value D, and determines the distance between adjacent dimensions. Is less than the threshold D, the mel LSP
Is performed to increase the distance between adjacent dimensions, and outputs the corrected mel LSP to the mel inverse transform unit 10.

Specifically, the mel LSP to be corrected is m
(K), where k = 1 to M, distance d (k) between adjacent dimensions
Is as shown below, and this distance d between adjacent dimensions is
When (k) falls below the threshold value D, as shown below, the mel LS
The P correction process is performed. Here, m ′ is the corrected mel LSP. d (k) = m (k + 1) -m (k) m '(k) = {m (k) + m (k + 1)} / 2-D / 2 m' (k + 1) = {m (k) + m (k + 1) )｝ / 2 + D / 2

The extension processing in the inter-dimension distance extension unit 9 is not limited to this, but is described in Japanese Patent Application Laid-Open No. 8-305.
Various methods such as the method disclosed in Japanese Patent No. 397 can be used. Further, the configuration of the LSP deforming unit 7 is not limited to the configuration of the second embodiment.

Then, upon receiving the corrected mel LSP from the LSP transforming section 7, the mel inverse transform section 10 receives the corrected mel LS
For each dimension value of P, an inverse transformation process for returning from the mel frequency domain to the linear frequency domain is performed, and a corrected LSP as a result of the inverse transformation is output. The conversion from the mel frequency domain to the linear frequency domain includes a method of calculating based on a table lookup and a method of calculating using a calculation formula corresponding to the inverse conversion of the approximate expression in the mel conversion unit 6.

As is clear from the above, this embodiment 2
According to the method described above, since the correction process is performed after converting the frequency domain of the input LSP from the linear frequency domain to the mel frequency domain, an effect that can provide an auditory equal effect over the entire frequency band is obtained. Play. Also, even when the audio encoding device and the audio decoding device are connected in multiple stages, or when there is a lot of background noise, without causing deterioration of the low frequency band,
There is an effect that a steep pole generated in a high frequency range can be satisfactorily corrected.

Note that LS using the approximate expression of the mel conversion unit 6
When the P correction device is applied to an audio encoding device or an audio decoding device, for example, the degradation of the high-frequency encoding characteristics tends to be large, or the low-frequency correction is weak and the amplitude of the decoded sound is low. In the case where the frequency becomes unstable, the approximation formula can be adjusted so as to convert to an intermediate frequency region between the linear frequency region and the mel frequency region. In this case, it is, of course, necessary to adjust the mel inverse transform unit 10 accordingly.

Embodiment 3 In the first and second embodiments, the LSP is corrected in the bar frequency domain or the mel frequency domain.
The conversion unit 1 is changed to a logarithmic frequency conversion unit (a conversion unit that converts the LSP frequency domain from a linear frequency domain to a logarithmic frequency domain), and the bark inverse conversion unit 5 is replaced with a logarithmic frequency inverse conversion unit (the LSP frequency domain (Inverse transform unit that returns to the frequency domain), and the LSP may be corrected in the logarithmic frequency domain. Further, in addition to the bark frequency domain, the mel frequency domain, or the logarithmic frequency domain, the LSP may be corrected by converting to a frequency domain having relatively good correspondence with auditory characteristics.

As is apparent from the above, the third embodiment
According to the method described above, since the correction process is performed after the frequency domain of the input LSP is converted from the linear frequency domain to the logarithmic frequency domain, an effect that can provide an auditory equal effect over the entire frequency band is obtained. Play. Also, even when the audio encoding device and the audio decoding device are connected in multiple stages, or when there is a lot of background noise, without causing deterioration of the low frequency band,
There is an effect that a steep pole generated in a high frequency range can be satisfactorily corrected.

Embodiment 4 FIG. 4 is a block diagram showing an LSP correction apparatus according to Embodiment 4 of the present invention. In FIG. 4, when an LSP which is a spectrum parameter is input, the distance between adjacent dimensions is determined based on each dimension value of the LSP. A threshold calculating unit (threshold calculating means) for calculating a threshold value for each dimension, a bark converting unit for converting an LSP frequency domain from a linear frequency domain to a bark frequency domain (frequency domain corresponding to auditory characteristics), 13 Is ba
bark whose frequency domain has been transformed by the rk transform unit 12
A temporary bank value calculating unit that adds and subtracts a half value of a predetermined threshold value D to each dimension value of the LSP, and outputs the addition result and the subtraction result as a temporary bark value;
3 is the frequency domain of the provisional bark value output from bark
A bark inverse transform unit that performs inverse transform from the frequency domain to the linear frequency domain and outputs the inverse transform result as a provisional frequency value;
Reference numeral 15 denotes a difference calculation unit that calculates a difference between provisional frequency values for each dimension output from the bark inverse transform unit 14 and outputs the calculation result as an inter-dimension distance threshold.

Reference numeral 16 denotes an LSP deformation unit (correction means) for correcting the LSP based on the threshold value for each dimension calculated by the threshold value calculation unit 11, 17 an inter-dimensional distance calculation unit for calculating the distance between adjacent dimensions of the LSP, Reference numeral 18 denotes an inter-dimension distance extension unit that increases the inter-dimension distance when the inter-dimension distance calculated by the inter-dimension calculation unit 17 is smaller than a threshold calculated from the inter-dimension distance threshold output from the threshold calculation unit 11. It is.

Next, the operation will be described. First, input L
The SP includes a bark converter 12 of the threshold calculator 11 and an LSP
It is input to the inter-dimension distance calculation unit 17 and the inter-dimension distance expansion unit 18 of the deformation unit 16. When the LSP is input, the bark conversion unit 12 converts each dimension value of the LSP into a bark frequency domain, and outputs a bark-LSP, which is a result of the conversion, as a temporary b.
Output to the ark value calculation unit 13.

Then, the provisional bark value calculating unit 13
When the bark-LSP is received from the rk conversion unit 12, a half value of a predetermined threshold value D is added to and subtracted from each dimension value of the bark-LSP as described below, and the addition result and the subtraction result are used as a temporary bark value. Output to the bark inverse transform unit 14. That is, for each dimension, two temporary b
Calculate and output the ark value. Temporary bark value = (each dimension value of bark-LSP) + D /
2 Temporary bark value = (bark-each dimension value of LSP)-D /
2

When the bark inverse transform unit 14 receives two temporary bark values for each dimension from the temporary bark value calculation unit 13, the bark inverse transform unit 14 inversely transforms the frequency domain of the temporary bark value from the bark frequency domain to the linear frequency domain. The result of the inverse conversion is output to the difference calculator 15 as a temporary frequency value.

When the difference calculating unit 15 receives two provisional frequency values for each dimension from the bark inverse transform unit 14,
A difference between two provisional frequency values is calculated for each dimension, and the calculation result is used as an inter-dimension distance threshold value for each dimension.
6 is output.

On the other hand, when the LSP transformation unit 16 receives the LSP, the interdimensional distance calculation unit 17 calculates the distance between adjacent dimensions of the LSP. Then, when the inter-dimensional distance calculating unit 17 calculates the distance between adjacent dimensions of the LSP, the inter-dimensional distance extending unit 18 of the LSP deforming unit 16 compares the distance between the adjacent dimensions of the LSP with the difference calculating unit 15 of the threshold calculating unit 11. Is compared with the threshold value calculated from the inter-dimensional distance threshold value output from. Then, when the distance between adjacent dimensions of the LSP is smaller than the threshold value, a process of correcting the corresponding dimension value of the input LSP and increasing the distance between adjacent dimensions is executed.

Specifically, the input LSP to be corrected is represented by f
(K), where k = 1 to M, distance d (k) between adjacent dimensions
Is as shown below, and this distance d between adjacent dimensions is
When (k) falls below a threshold value D '(k) calculated from the inter-dimension distance threshold value Df (k) for each dimension, the input LSP is corrected as described below. Here, f ′ is the corrected input LSP. d (k) = f (k + 1) -f (k) D'f (k) = (Df (k) + Df (k + 1)) / 2 f '(k) = {f (k) + f (k + 1)} / 2-D'f (k) f '(k + 1) = {f (k) + f (k + 1)} / 2 + D'f (k)

The extension processing in the inter-dimension distance extension unit 18 is not limited to this, but is described in
Various methods such as the method disclosed in Japanese Patent No. 5397 can be used. Further, the configuration of LSP deforming section 16 is not limited to the configuration of the fourth embodiment.

FIG. 5 is an explanatory diagram for explaining a correction result of the LSP correction device according to the fourth embodiment. FIG. 5A shows the k-th bar-LSP output from the bark conversion unit 12.
The following values b (k) and corresponding temporary bark values b _t1 (k) and b _t2 (k) output by the temporary bark value calculation unit 13 are shown. The temporary bark value is calculated by adding and subtracting a half value of a predetermined threshold D to each dimension value of the bark-LSP.
FIG. 15B shows the provisional bark value of FIG.
Temporary frequency value f output by inverse conversion by inverse conversion unit 14
_t1 (k) and _ft2 (k) are shown. Difference calculator 15
Calculates the k-th dimension distance threshold Df (k) by taking the difference between these two provisional frequency values.

As is apparent from the above, the fourth embodiment
According to the method, the threshold value for the distance between adjacent dimensions is calculated for each dimension according to the value of each dimension of the input LSP, and the input LSP is corrected based on the threshold value. The input LSP is corrected based on the optimal threshold value, and as a result, the influence of the correction is different for each frequency, and the coding / decoding quality of the applied voice coding apparatus and voice decoding apparatus is reduced. Deterioration cannot be eliminated,
On the contrary, there is an effect that the problem of causing deterioration can be solved.

Further, since the threshold value is calculated in the bark frequency region having a good correspondence to the auditory characteristics, an effect that an auditory equal effect can be obtained over the entire frequency band is obtained. Furthermore, even when the speech encoding device and speech decoding device are connected in multiple stages, or when there is much background noise,
There is an effect that a steep pole generated in a high frequency can be favorably corrected without causing deterioration in a low frequency.

When an LSP correction device using the approximation formula of the bark conversion unit 12 is applied to a speech coding device or a speech decoding device, for example, the degradation of high-frequency coding characteristics tends to be large. On the other hand, if the low-frequency correction is weak and the amplitude of the decoded sound becomes unstable, the approximation formula may be adjusted to convert to an intermediate frequency domain between the linear frequency domain and the bark frequency domain. it can. In this case, the bark inverse transform unit 14 also needs to be adjusted accordingly.

Embodiment 5 FIG. 6 is a block diagram showing an LSP correction apparatus according to Embodiment 5 of the present invention. In the figure, the same reference numerals as those in FIG. 4 denote the same or corresponding parts, and a description thereof will be omitted. 21 is a spectrum parameter L
When an SP is input, a threshold calculator (threshold calculating means) for calculating a threshold for the distance between adjacent dimensions for each dimension based on each dimension value of the LSP. A mel transform unit 23 for transforming into a frequency domain (frequency domain corresponding to auditory characteristics), and a mel LSP 23 whose frequency domain is transformed by the mel transform unit 22
A temporary mel value calculator that adds and subtracts a half value of a predetermined threshold value D to and from each dimension value and outputs the addition result and the subtraction result as a temporary mel value. A mel inverse transform unit that inversely transforms the frequency domain of the value from the mel frequency domain to the linear frequency domain and outputs the result of the inverse transform as a temporary frequency value. This is a difference calculation unit that calculates the difference between the frequency values and outputs the calculation result as a distance threshold between dimensions.

Next, the operation will be described. First, input L
The SP is the mel conversion unit 22 of the threshold value calculation unit 21, the inter-dimension distance calculation unit 17 of the LSP deformation unit 16, and the inter-dimension distance expansion unit 18
Is input to Upon input of the LSP, the mel transform unit 22 converts each dimension value of the LSP into a mel frequency domain,
The mel LSP as a result of the conversion is output to the provisional mel value calculation unit 23.

Upon receiving the mel LSP from the mel conversion unit 22, the temporary mel value calculation unit 23 calculates
A half value of a predetermined threshold value D is added to and subtracted from each dimension value of the mel LSP, and the addition result and the subtraction result are output to the mel inverse conversion unit 24 as a temporary mel value. That is, for each dimension, two temporary mel values are calculated and output based on the dimension value. Temporary Mel Value = Each Dimension Value of Mel LSP + D / 2 Temporary Mel Value = Each Dimension Value of Mel LSP−D / 2

The inverse mel conversion unit 24 is provided with a temporary mel value calculation unit 23.
When two temporary mel values are received for each dimension from, the frequency domain of the tentative mel value is inversely transformed from the mel frequency domain to the linear frequency domain, and the result of the inverse transformation is used as a tentative frequency value in the difference calculation unit 2.
5 is output.

Upon receiving the two provisional frequency values for each dimension from the mel inverse transform unit 24, the difference calculation unit 25 calculates the difference between the two provisional frequency values for each dimension, and calculates the result. Is output to the LSP transformation unit 16 as an inter-dimensional distance threshold for each dimension. Subsequent operations of the LSP deforming unit 16 are the same as those of the fourth embodiment, and thus description thereof is omitted.

As is apparent from the above, the fifth embodiment
According to the method, the threshold value for the distance between adjacent dimensions is calculated for each dimension according to the value of each dimension of the input LSP, and the input LSP is corrected based on the threshold value. The input LSP is corrected based on the optimal threshold value, and as a result, the influence of the correction is different for each frequency, and the coding / decoding quality of the applied voice coding apparatus and voice decoding apparatus is reduced. Deterioration cannot be eliminated,
On the contrary, there is an effect that the problem of causing deterioration can be solved.

Further, since the threshold value is calculated in the mel frequency region having a good response to the auditory characteristics, an effect that an auditory equal effect can be obtained over the entire frequency band is achieved. Furthermore, even when the speech encoding device and the speech decoding device are connected in multiple stages or there is a lot of background noise, it is possible to satisfactorily correct a steep pole generated in a high frequency band without causing deterioration in a low frequency region. It works.

Note that L using the approximate expression of the mel conversion unit 22
When the SP correction device is applied to a voice coding device or a voice decoding device, for example, the degradation of the high-frequency coding characteristics tends to be large, or the low-frequency correction is weak and the amplitude of the decoded sound is low. In the case where the frequency becomes unstable, the approximation formula can be adjusted so as to convert to an intermediate frequency region between the linear frequency region and the mel frequency region. In this case, it is necessary to adjust the mel inverse transform unit 24 accordingly.

Embodiment 6 FIG. In the above fourth and fifth embodiments, the calculation of the threshold by the calculation in the bark frequency domain or the mel frequency domain has been described. However, the bark transform unit 12 in FIG. To a logarithmic frequency domain), the temporary bark value calculator 13 is changed to a temporary logarithmic frequency calculator (calculator for calculating a temporary logarithmic frequency value), and the bark inverse converter 14 The threshold may be calculated by changing to a logarithmic frequency inverse transform unit (an inverse transform unit for returning the frequency domain of the LSP to the linear frequency domain) and calculating in the logarithmic frequency domain. In addition, the threshold may be calculated by calculation in a frequency domain having relatively good correspondence with auditory characteristics, in addition to the bark frequency domain, the mel frequency domain, or the logarithmic frequency domain.

Embodiment 7 FIG. FIG. 7 is a block diagram showing an LSP correction apparatus according to Embodiment 7 of the present invention. In the figure, the same reference numerals as those in FIG. 4 denote the same or corresponding parts, and a description thereof will be omitted. Reference numeral 31 denotes a maximum value selection unit that compares the inter-dimension distance threshold calculated for each dimension by the difference calculation unit 15 with a fixed threshold defined in the linear frequency domain, and outputs the larger threshold to the LSP transformation unit 16. is there.

In the fourth embodiment, the inter-dimension distance threshold for each dimension calculated by the difference calculator 15 is always set to the LSP.
Although the output to the transformation unit 16 is shown, the maximum value selection unit 31 compares the inter-dimension distance threshold calculated for each dimension by the difference calculation unit 15 with the fixed threshold defined in the linear frequency domain, and The other threshold may be output to the LSP transformation unit 16.

According to the seventh embodiment, in accordance with the value of each dimension of the input LSP, the threshold value for the distance between adjacent dimensions is calculated in the frequency domain having a good correspondence to the auditory characteristics. Since the maximum value among the threshold values defined in is set as the final threshold value, in addition to the effect of the fourth embodiment, an acoustically uniform effect is provided over the entire frequency band, and And an LSP correction device that does not output an unstable LSP (a state in which the inter-dimensional distance is too small in a low frequency range and the filter gain of the LSP is too large) is provided.

In the seventh embodiment, the case where the maximum value selector 31 is applied to the LSP correction device of the fourth embodiment has been described.
The maximum value selection unit 31 may be applied to the correction device.

Embodiment 8 FIG. FIG. 8 is a configuration diagram showing an LSP correction apparatus according to Embodiment 8 of the present invention. In the figure, the same reference numerals as those in FIG. 4 denote the same or corresponding parts, and a description thereof will not be repeated. 32 is a spectrum parameter L
When an SP is input, two frequency values close to each dimension value (frequency value) of the LSP are searched from the frequency-to-threshold table 33, and a threshold calculation for calculating an inter-dimensional distance threshold of each dimension based on the two frequency values is performed. Unit (threshold value calculating means), 33 is a frequency-to-threshold table for storing a representative frequency value and a threshold value corresponding to the representative frequency value, and 34 is for searching for two threshold values corresponding to the representative frequency value. And an interpolation unit for interpolating two threshold values.

Next, the operation will be described. In the above fourth to seventh embodiments, the calculation of the threshold by the calculation in the bark frequency domain or the mel frequency domain has been described.
Two frequency values close to each dimension value of the input LSP may be searched from the frequency-threshold table 33, and the inter-dimension distance threshold of each dimension may be calculated based on the two frequency values.

More specifically, first, the interpolator 34 of the threshold calculator 32 searches the frequency-threshold table 33 for two frequency values close to each dimension value (frequency value) of the input LSP.
Then, when the interpolation unit 34 searches for two frequency values,
Two threshold values corresponding to the two frequency values are obtained from the frequency-threshold table 33.

Then, the interpolation unit 34 performs weighted addition of the two thresholds according to the difference between the two frequency values and the order value in the input LSP, and calculates the dimension distance threshold of the order. This calculation process is executed for each order, and the dimension distance threshold for each dimension is output to the dimension extension section 18 of the LSP deforming section 16. Subsequent operations of the LSP deforming unit 16 are the same as those of the fourth embodiment, and thus description thereof is omitted.

As is clear from the above, this embodiment 8
According to the method, the threshold value for the distance between adjacent dimensions is calculated for each dimension according to the value of each dimension of the input LSP, and the input LSP is corrected based on the threshold value. The input LSP is corrected based on the optimal threshold value, and as a result, the influence of the correction is different for each frequency, and the coding / decoding quality of the applied voice coding apparatus and voice decoding apparatus is reduced. Deterioration cannot be eliminated,
On the contrary, there is an effect that the problem of causing deterioration can be solved.

Further, even when the speech encoding device and the speech decoding device are connected in multiple stages or when there is a lot of background noise, the steep pole generated in the high frequency range is corrected well without causing deterioration in the low frequency range. The effect that can be achieved. Although a new memory for the table is required as compared with the fourth embodiment, the effect of simplifying the processing is obtained.

Embodiment 9 FIG. FIG. 9 is a block diagram showing a speech encoding apparatus according to Embodiment 9 of the present invention. In the figure, the same reference numerals as those in FIG. 1 denote the same or corresponding parts, and a description thereof will be omitted. 41 is an LSP analysis unit (LSP analysis means) for analyzing an input voice to calculate an LSP, and 42 is an LSP
An LSP correction device for correcting the LSP calculated by the analysis unit 41, and a correction LS 43 output by the LSP correction device 42
L that encodes P and outputs an LSP code and a quantized LSP
An SP encoding unit (LSP encoding unit) 44 is an excitation encoding unit (excitation encoding unit) that calculates an encoded excitation (excitation code) from the quantized LSP output from the LSP encoding unit 43 and the input speech. is there.

Next, the operation will be described. First, LSP
When the input voice is input, the analysis unit 41 analyzes the input voice, calculates an LSP, and outputs the LSP to the LSP correction device 42. In general, the LSP is calculated by performing a linear prediction analysis, calculating a linear prediction coefficient, and converting the coefficient.

When receiving the LSP from the LSP analysis unit 41, the LSP correction unit 42 corrects the LSP in the same procedure as in the first embodiment, and converts the corrected LSP to the LSP.
Output to the SP encoder 43. Then, when the LSP correction device 42 outputs the corrected LSP, the LSP encoding unit 43 encodes the corrected LSP and outputs an LSP code. For example, the LSP encoding unit 43 obtains a quantized LSP ( (The quantized LSP is equivalent to the result of encoding and decoding.)

[0096] Excitation encoding section 44 includes LSP encoding section 43.
Receives the quantized LSP from the source codec, it encodes the sound source information of the input voice using the quantized LSP, and outputs the coding result as a sound source code.

In the ninth embodiment, the LSP correction apparatus of the first embodiment is applied. However, the LSP correction apparatuses of the second to eighth embodiments may be applied. Alternatively, the input of the LSP analysis unit 41 may be changed to a decoded speech, and a backward CELP system configuration as disclosed in Japanese Patent Application Laid-Open No. Hei 5-273997 may be employed.

As is apparent from the above, the ninth embodiment
According to the LSP calculated by analyzing the input voice,
Since the correction device 42 corrects and encodes the corrected LSP, there is an effect that the quality degradation due to the distortion of the input voice and the analysis error can be suppressed well.

In particular, since the LSP correction device 42 corrects the LSP, it becomes possible to calculate the optimum threshold value for the frequency of each dimension of the LSP, and as a result, the effect of the correction is different for each frequency. There is an effect that it is possible to solve the problem that the degradation of the encoding / decoding quality cannot be eliminated or the degradation is caused. In addition, if the LSP correction process is performed after converting to a frequency region having good correspondence to the auditory characteristics, LSP correction that provides an auditory equal effect over the entire frequency band is realized. This provides an effect that good coding characteristics can be obtained. Furthermore, even when the speech encoding device and the corresponding speech decoding device are connected in multiple stages or when there is a lot of background noise, the steep poles generated in the high frequency range can be satisfactorily corrected without causing deterioration in the low frequency range. As a result, there is an effect that a good encoding characteristic can be obtained.

Embodiment 10 FIG. FIG. 10 is a configuration diagram showing a speech encoding apparatus according to Embodiment 10 of the present invention.
9, the same reference numerals as those in FIG. 9 denote the same or corresponding parts, and a description thereof will not be repeated. Reference numeral 45 denotes L for encoding the LSP calculated by the LSP analysis unit 41 and outputting an LSP code
The SP encoder (LSP encoder) 46 decodes the LSP code output from the LSP encoder 45 to decode LSP code.
An LSP decoding unit that outputs SP (LSP decoding means);
Reference numeral 47 denotes an LSP correction device for correcting the decoded LSP output from the LSP decoding unit 46, and reference numeral 48 denotes a sound source coding unit (calculated from the decoded LSP corrected by the LSP correction device 47 and the input voice, which calculates an encoded sound source (sound source code)). (Excitation coding means).

Next, the operation will be described. First, LSP
When inputting the input voice, the analysis unit 41 analyzes the input voice and calculates the LSP, as in the ninth embodiment.
The LSP is output to the LSP encoder 45.

Then, when receiving the LSP from the LSP analyzer 41, the LSP encoder 45 encodes the LSP and outputs an LSP code. Then, the LSP decoding unit 46
When the LSP encoding unit 45 outputs the LSP code, the LSP encoding unit 45 decodes the LSP code and outputs the decoded LSP to the LSP correction device 47.

The LSP correction unit 47 includes an LSP decoding unit 4
When the decoding LSP is received from 6, the decoding LSP is corrected in the same procedure as in the first embodiment, and the corrected LSP is output to excitation coding section 48. Then, excitation coding section 4
8 receives the correction LSP from the LSP correction device 47,
The sound source information of the input voice is encoded using the corrected LSP, and the encoded result is output as a sound source code.

In the tenth embodiment, the LSP correction apparatus of the first embodiment is applied. However, the LSP correction apparatuses of the second to eighth embodiments may be applied.

As is apparent from the above, the first embodiment
According to 0, the LSP correction device 47 corrects the decoded LSP and encodes the excitation information using the corrected LSP, so that the decoded sound becomes unstable due to LSP coding distortion (quantization distortion). This has the effect of successfully suppressing the formation.

In particular, since the LSP correction device 47 corrects the LSP, it becomes possible to calculate the optimum threshold value for the frequency of each dimension of the LSP, and as a result, the influence of the correction is different for each frequency. There is an effect that it is possible to solve the problem that the degradation of the encoding / decoding quality cannot be eliminated or the degradation is caused. In addition, if the LSP correction process is performed after converting to a frequency region having good correspondence to the auditory characteristics, LSP correction that provides an auditory equal effect over the entire frequency band is realized. This provides an effect that good coding characteristics can be obtained. Furthermore, even when the speech encoding device and the corresponding speech decoding device are connected in multiple stages or when there is a lot of background noise, the steep poles generated in the high frequency range can be satisfactorily corrected without causing deterioration in the low frequency range. As a result, there is an effect that a good encoding characteristic can be obtained.

Embodiment 11 FIG. FIG. 11 is a configuration diagram showing a speech decoding apparatus according to Embodiment 11 of the present invention.
In the figure, the same reference numerals as those in FIG. 51 is an LSP decoding unit (LSP decoding means) for decoding the LSP code and outputting a decoded LSP, 52 is a decoded LSP output from the LSP decoding unit 51
A sound source decoding unit (sound source decoding means) for decoding a sound source code to generate a sound source signal, and a correction LSP output from the LSP correction device 52.
And a synthesis filter (synthesis unit) for generating a synthesized sound from the sound source signal generated by the sound source decoding unit 53.

Next, the operation will be described. First, LSP
When the decoding unit 51 receives the LSP code, the LSP code
The code is decoded and the decoded LSP is output to the LSP correction device 52. This LSP code is output from the speech coding device or the like described in the tenth embodiment.

The LSP correction device 52 includes an LSP decoding unit 5
When receiving the decoded LSP from the first LSP, the decoding LSP is corrected in the same procedure as in the first embodiment, and the corrected LSP is output to the synthesis filter 54. On the other hand, the sound source decoding unit 53
Receives the excitation code, generates the excitation signal by decoding the excitation code, and outputs the excitation signal to the synthesis filter 54.

When receiving the corrected LSP from the LSP corrector 52, the synthesis filter 54 performs a synthesis filtering process on the sound source signal using the corrected LSP, and uses the resulting synthesized sound as an output voice. Output. In the eleventh embodiment, the LSP correction device of the first embodiment is applied. However, the LSP correction device of the second to eighth embodiments may be applied.

As is clear from the above, the first embodiment
According to No. 1, since the LSP correction device 52 corrects the decoded LSP and performs the synthesis filtering process using the corrected LSP, it is possible to appropriately suppress the instability of the decoded sound due to the transmission error of the LSP code. It has the effect of being able to. Further, the LSP code output from the speech encoding device or the like according to the tenth embodiment is input and the LS
When the P correction device 52 is the same as the LSP correction device of the speech coding device, the LSP coding distortion (quantization distortion)
Thus, it is possible to provide an audio decoding device that can appropriately suppress the instability of the decoded sound due to.

In particular, since the LSP correction device 52 corrects the decoded LSP, it becomes possible to calculate the optimum threshold value for the frequency of each dimension of the decoded LSP.
There is an effect that it is possible to solve the problem that the influence of the correction is different for each frequency and the degradation of the encoding / decoding quality cannot be eliminated, or the degradation of the quality can be solved. In addition, if the LSP correction process is performed after converting to a frequency region having good correspondence to the auditory characteristics, LSP correction that provides an auditory equal effect over the entire frequency band is realized. Thus, there is an effect that good decoding characteristics can be obtained. Further, even if the speech encoding device corresponding to the speech decoding device and the speech decoding device are connected in multiple stages, or if there is much background noise input to the speech encoding device, low-frequency degradation is caused. Without this, it is possible to satisfactorily correct a steep pole generated in a high frequency range and obtain an effect of obtaining a good decoding characteristic.

Embodiment 12 FIG. FIG. 12 is a configuration diagram showing a speech decoding apparatus according to Embodiment 12 of the present invention.
In the figure, the same reference numerals as those in FIG. 11 denote the same or corresponding parts, and a description thereof will not be repeated. Reference numeral 55 denotes a synthesis filter (synthesis unit) that generates a synthesized sound from the decoded LSP output from the LSP decoding unit 51 and the sound source signal generated by the sound source decoding unit 53, and reference numeral 56 denotes a corrected LSP output from the LSP correction device 52. Is a post-filter (post-filter means) for executing a spectrum emphasis process on a synthesized sound by using.

Next, the operation will be described. First, LSP
When the decoding unit 51 receives the LSP code, the LSP code
The code is decoded and the decoded LSP is output to the LSP correction device 52 and the synthesis filter 55. This LSP code is output from the speech coding device or the like described in the tenth embodiment.

On the other hand, when excitation codec 53 receives the excitation code, it decodes the excitation code to generate an excitation signal and outputs the excitation signal to synthesis filter 55. Then, when receiving the decoded LSP from the LSP decoding unit 51, the LSP correction device 52 corrects the decoded LSP in the same procedure as in the first embodiment, and outputs the corrected LSP to the post filter 56.

When receiving the decoded LSP from the LSP decoding unit 51, the synthesis filter 55 executes synthesis filtering processing on the sound source signal using the decoded LSP, and converts the resultant synthesized sound into a post-filter. 5
6 is output. And the post filter 56 is an LSP
When receiving the correction LSP from the correction device 52, the correction LS
A voice emphasizing process is performed on the synthesized sound using P, and the processed synthesized sound obtained as a result is output as an output sound.

In the twelfth embodiment, the LSP correction apparatus of the first embodiment is applied. However, the LSP correction apparatuses of the second to eighth embodiments may be applied. Further, in combination with the eleventh embodiment, the correction LSP may also be input to the synthesis filter 55.

As is clear from the above, the first embodiment
According to 2, since the post-filter processing is performed using the correction LSP, it is possible to prevent the pole of the post-filter from being too steep to cause over-emphasis. Further, since the LSP correction device 52 performs the correction, it is possible to calculate the optimum threshold value for the frequency of each dimension of the decoded LSP, and as a result, it is possible to obtain an effect of obtaining good voice enhancement.

Further, when the LSP correction process is performed after converting to a frequency region having a good correspondence to the auditory characteristics, the LSP correction that provides an auditory equal effect over the entire frequency band is realized. Therefore, there is an effect that a good voice emphasis characteristic can be obtained. Further, even if the speech encoding device corresponding to the speech decoding device and the speech decoding device are connected in multiple stages, or if there is much background noise input to the speech encoding device, low-frequency degradation is caused. Without this, it is possible to satisfactorily correct a steep pole generated in a high frequency range and obtain an effect of obtaining a good voice emphasis characteristic.

Embodiment 13 FIG. FIG. 13 is a block diagram showing a post filter in an audio decoding apparatus according to Embodiment 13 of the present invention. In the figure, reference numerals 61 and 62 denote LSP correction apparatuses for correcting a decoded LSP, and reference numerals 63 and 64 denote LSPs.
An LPC conversion unit for converting the corrected LSP output from the correction devices 61 and 62 into an LPC area, and a synthesis filter process 65 for the synthesized sound separately generated in the speech decoding device using the LPC output from the LPC conversion unit 63 LPC to execute
The synthesis filter 66 is an LP output from the LPC conversion unit 64
This is an LPC inverse filter that performs inverse filtering on the output signal of the LPC synthesis filter 65 using C.
Note that the overall configuration of the speech decoding device may be the same as that of FIG. 12 or may be another configuration.

Next, the operation will be described. First, decryption L
The SP is input to the LSP correction devices 61 and 62. The decoded LSP is generated in the audio decoding device and is the same as or different from the one used for the synthesis filter in the audio decoding device. Is corrected in the same manner as in the twelfth embodiment.

When receiving the decoded LSP, the LSP correction device 61 corrects the decoded LSP and outputs the corrected LSP to the LPC conversion unit 63. Then, the LPC conversion unit 6
3 receives the correction LSP from the LSP correction device 61,
The corrected LSP is converted into an LPC area, and the resulting LPC is output to the LPC synthesis filter 65.

On the other hand, the LSP correction device 62 outputs the decrypted LSP
Is input, the decoded LSP is corrected, and the corrected LSP is corrected.
The SP is output to the LPC conversion unit 64. Then, when receiving the correction LSP from the LSP correction device 62, the LPC conversion unit 64 converts the correction LSP into an LPC area, and outputs the LPC obtained as a result to the LPC inverse filter 66.

Then, the LPC synthesis filter 65
When the C conversion unit 63 outputs the LPC, the LPC is used to perform a synthesis filter process on a synthesized sound separately generated in the speech decoding apparatus, and the resultant signal is converted to an L signal.
Output to the PC inverse filter 66.

The LPC inverse filter 66 is an LPC inverse filter.
Using the LPC output from the conversion unit 64, an inverse filter process is performed on the output signal of the LPC synthesis filter 65,
The signal obtained as a result is output as processed synthetic sound. In some cases, the processed synthesized sound is the final output sound of the entire speech decoding device, and in another case, the processed sound is further processed in a later stage to be the final output sound.

In the thirteenth embodiment, the LSP correction apparatus of the first embodiment is applied. However, the LSP correction apparatuses of the second to eighth embodiments may be applied.

As is apparent from the above, the first embodiment
According to No. 3, since the LSP correction devices 61 and 62 correct the decoded LSP and perform the post-filter processing using the corrected LSP, it is possible to calculate the optimum threshold value for the frequency of each dimension of the decoded LSP. As a result, it is possible to obtain an effect that good voice enhancement can be obtained. In addition, after converting to a frequency domain that is compatible with auditory characteristics, LSP
When the correction processing is performed, the LSP correction that provides an equal effect perceptually over the entire frequency band is realized, so that an effect that a good voice emphasis characteristic is obtained can be obtained. Furthermore, a speech encoding device corresponding to the speech decoding device and the speech decoding device are connected in multiple stages,
Even if there is much background noise input to the speech coding device,
Without causing deterioration in the low frequency range, steep poles generated in the high frequency range are satisfactorily corrected, and an effect of obtaining good voice emphasis characteristics can be obtained.

[0128]

As described above, according to the present invention, the LSP converted into the frequency domain corresponding to the auditory characteristic is corrected, and the frequency domain of the LSP is returned to the linear frequency domain. There is an effect that an auditory equal effect can be provided over the entire frequency band.

According to the present invention, since the LSP frequency domain is converted from the linear frequency domain to the Bark frequency domain for correction, it is possible to provide an auditory equal effect over the entire frequency band. effective.

According to the present invention, since the LSP frequency domain is converted from the linear frequency domain to the mel frequency domain for correction, it is possible to provide an acoustically uniform effect over the entire frequency band. effective.

According to the present invention, since the LSP frequency domain is converted from the linear frequency domain to the logarithmic frequency domain and corrected, it is possible to provide an acoustically uniform effect over the entire frequency band. effective.

According to the present invention, the threshold for the distance between adjacent dimensions is calculated for each dimension based on each dimension value of the LSP, and the LSP is corrected based on the threshold for each dimension. To solve the problem that the influence of the correction is different for each frequency, and the deterioration of the coding / decoding quality of the applied speech coding apparatus and the speech decoding apparatus cannot be eliminated or, conversely, causes the degradation. There is an effect that can be.

According to the present invention, the LSP frequency domain is converted from the linear frequency domain to the frequency domain corresponding to the auditory characteristics, and the threshold value is calculated in the frequency domain corresponding to the auditory characteristics. There is an effect that an auditory equal effect can be provided over the entire frequency band.

According to the present invention, the threshold calculated in the frequency domain corresponding to the auditory characteristic is compared with the threshold defined in the linear frequency domain, and the larger threshold is output to the correction means. An LSP that produces an auditory equal effect over the entire frequency band and does not output an unstable LSP (a state in which the filter gain of the LSP becomes too large due to the inter-dimensional distance being too close in the low frequency range). There is an effect that the correction device can provide.

According to the present invention, since the threshold is calculated by converting the LSP frequency domain from the linear frequency domain to the Bark frequency domain, it is possible to provide an auditory equal effect over the entire frequency band. There is an effect that can be.

According to the present invention, since the LSP frequency domain is converted from the linear frequency domain to the mel frequency domain to calculate the threshold value, an auditory equal effect can be obtained over the entire frequency band. There is an effect that can be.

According to the present invention, the threshold is calculated by converting the frequency domain of the LSP from the linear frequency domain to the logarithmic frequency domain, so that an auditory equal effect is provided over the entire frequency band. There is an effect that can be.

According to the present invention, since the LSP corrected in the frequency domain corresponding to the auditory characteristics is encoded and the LSP code and the quantized LSP are output, the quality of the input speech due to distortion and analysis error is increased. There is an effect that deterioration can be favorably suppressed.

According to the present invention, the LSP corrected using the threshold value for each dimension calculated based on each dimension value of the LSP.
Since the SP is encoded and the LSP code and the quantized LSP are output, there is an effect that the quality deterioration due to the distortion and the analysis error of the input voice can be suppressed well.

According to the present invention, since the coded excitation is calculated from the decoded LSP corrected in the frequency domain corresponding to the auditory characteristics and the input voice, the coding distortion of the LSP is reduced.
There is an effect that the instability of the decoded sound due to (quantization distortion) can be favorably suppressed.

According to the present invention, the coded excitation is calculated from the decoded LSP and the input speech which are corrected using the threshold value for each dimension calculated based on each dimension value of the LSP. Thus, there is an effect that the instability of the decoded sound due to the encoding distortion (quantization distortion) can be favorably suppressed.

According to the present invention, since the synthesized sound is generated from the decoded LSP corrected in the frequency domain corresponding to the auditory characteristics and the sound source signal generated by the sound source decoding means, the transmission error of the LSP code can be improved. Thus, there is an effect that the instability of the decoded sound due to the above can be suppressed well.

According to the present invention, a synthesized sound is generated from the decoded LSP corrected by using the threshold value for each dimension calculated based on each dimension value of the LSP and the sound source signal generated by the sound source decoding means. With such a configuration, there is an effect that the instability of the decoded sound due to the transmission error of the LSP code can be favorably suppressed.

According to the present invention, the spectrum emphasis processing is performed on the synthesized sound using the decoded LSP corrected in the frequency domain corresponding to the auditory characteristics.
There is an effect that good voice enhancement can be obtained while suppressing that the pole of the post-filter is too steep to cause over-emphasis.

According to the present invention, the spectrum emphasis processing for the synthesized sound is performed using the decoded LSP corrected using the threshold value for each dimension calculated based on each dimension value of the LSP. In addition, there is an effect that good voice enhancement can be obtained while suppressing that the pole of the post filter is too steep to cause over-emphasis.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing an LSP correction device according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram illustrating a correction result of the LSP correction device according to the first embodiment.

FIG. 3 is a configuration diagram showing an LSP correction device according to a second embodiment of the present invention.

FIG. 4 is a configuration diagram showing an LSP correction device according to a fourth embodiment of the present invention.

FIG. 5 is an explanatory diagram illustrating a correction result of the LSP correction device according to the fourth embodiment.

FIG. 6 is an explanatory diagram illustrating a correction result of the LSP correction device according to the fifth embodiment.

FIG. 7 is an explanatory diagram illustrating a correction result of the LSP correction device according to the seventh embodiment.

FIG. 8 is an explanatory diagram illustrating a correction result of the LSP correction device according to the eighth embodiment.

FIG. 9 is a configuration diagram illustrating a speech encoding device according to a ninth embodiment.

FIG. 10 is a configuration diagram illustrating a speech encoding device according to a tenth embodiment.

FIG. 11 is a configuration diagram illustrating a speech decoding apparatus according to an eleventh embodiment.

FIG. 12 is a configuration diagram illustrating a speech decoding apparatus according to a twelfth embodiment.

FIG. 13 is a configuration diagram showing a post filter in a speech decoding apparatus according to Embodiment 13;

[Explanation of symbols]

1 bark conversion unit (conversion means), 2, 7, 16 LS
P transformation unit (correction unit), 5 bark inverse conversion unit (inversion unit), 6 mel conversion unit (conversion unit), 10 mel inverse conversion unit (inverse conversion unit), 11, 21, 32 Threshold calculation unit (threshold calculation) Means), 41 LSP analysis section (LSP analysis means), 43, 45 LSP encoding section (LSP encoding means), 44, 48 excitation encoding section (excitation encoding means),
46, 51 LSP decoding section (LSP decoding means), 53
Sound source decoding unit (sound source decoding means), 54, 55 synthesis filter (synthesis means), 56 post filter (post filter means).

Claims (18)

[Claims] When an LSP which is a spectrum parameter is input, a conversion means for converting a frequency domain of the LSP from a linear frequency domain to a frequency domain corresponding to an auditory characteristic, and an LSP whose frequency domain is converted by the conversion means
An LSP correction apparatus comprising: a correction unit that corrects the LSP; 2. The LSP correction device according to claim 1, wherein the conversion means converts the LSP frequency domain from a linear frequency domain to a Bark frequency domain. 3. The LSP correction apparatus according to claim 1, wherein the conversion means converts the frequency domain of the LSP from a linear frequency domain to a mel frequency domain. 4. The LSP correction device according to claim 1, wherein the conversion means converts the frequency domain of the LSP from a linear frequency domain to a logarithmic frequency domain. 5. When a LSP which is a spectrum parameter is input, a threshold value calculating means for calculating a threshold value for a distance between adjacent dimensions for each dimension based on each dimension value of the LSP, and a threshold value calculating means for calculating a threshold value. Correction means for correcting the LSP based on a threshold value for each dimension. 6. The threshold value calculating means converts a frequency domain of an LSP from a linear frequency domain to a frequency domain corresponding to an auditory characteristic when calculating a threshold value regarding a distance between adjacent dimensions, and calculates a frequency corresponding to the auditory characteristic. The LSP correction device according to claim 5, wherein a threshold value is calculated in the area. 7. The threshold calculating means compares a threshold calculated in a frequency domain corresponding to an auditory characteristic with a threshold defined in a linear frequency domain, and outputs a larger threshold to the correcting means. The LSP correction device according to claim 6, wherein 8. The LSP correction device according to claim 6, wherein the threshold value calculating means calculates the threshold value by converting the frequency domain of the LSP from the linear frequency domain to the Bark frequency domain. 9. The LSP correction device according to claim 6, wherein the threshold value calculating means calculates the threshold value by converting a frequency domain of the LSP from a linear frequency domain to a mel frequency domain. 10. The LSP correction device according to claim 6, wherein the threshold value calculating means calculates the threshold value by converting a frequency domain of the LSP from a linear frequency domain to a logarithmic frequency domain. 11. An LSP analyzer for analyzing an input voice to calculate an LSP, and a converter for converting a frequency domain of the LSP calculated by the LSP analyzer from a linear frequency domain to a frequency domain corresponding to auditory characteristics. Correction means for correcting the LSP whose frequency domain has been converted by the conversion means, inverse conversion means for returning the frequency domain of the LSP corrected by the correction means to the linear frequency domain, and output from the inverse conversion means. LSP encoding means for encoding the LSP and outputting an LSP code and a quantized LSP;
An audio encoding device comprising: a quantized LSP output from a P encoding unit; and an excitation encoding unit that calculates an encoded excitation from the input audio. 12. An LSP analyzing means for analyzing an input voice to calculate an LSP, and a threshold value for a distance between adjacent dimensions is calculated for each dimension based on each dimension value of the LSP calculated by the LSP analyzing means. Threshold calculation means, correction means for correcting the LSP based on the threshold value for each dimension calculated by the threshold calculation means, LSP code corrected by the correction means, and an LSP code and quantization LS
An audio encoding apparatus comprising: an LSP encoding unit that outputs P; and an excitation encoding unit that calculates an encoded excitation from the quantized LSP output from the LSP encoding unit and the input audio. 13. An LSP analyzing unit for analyzing an input voice to calculate an LSP, an LSP encoding unit for encoding the LSP calculated by the LSP analyzing unit and outputting an LSP code, and an LSP encoding unit. Output LSP
LSP decoding means for decoding a code and outputting a decoded LSP, and a decoded LSP output from the LSP decoding means
Transforming the frequency domain from the linear frequency domain to the frequency domain corresponding to the auditory characteristic, correcting means for correcting the decoded LSP whose frequency domain has been converted by the converting means, and decoding corrected by the correcting means. A speech encoding apparatus comprising: an inverse transform unit for returning a frequency domain of an LSP to a linear frequency domain; and a sound source encoding unit for calculating an encoded excitation from the decoded LSP output from the inverse transform unit and the input speech. 14. An LSP analyzing means for analyzing an input voice to calculate an LSP, an LSP encoding means for encoding the LSP calculated by the LSP analyzing means and outputting an LSP code, and Output LSP
LSP decoding means for decoding a code and outputting a decoded LSP, and a decoded LSP output from the LSP decoding means
A threshold value calculating means for calculating a threshold value for the distance between adjacent dimensions for each dimension with reference to each dimension value of, and an LS value based on the threshold value for each dimension calculated by the threshold value calculating means.
A speech coding apparatus comprising: a correction unit for correcting P; a decoding LSP corrected by the correction unit; and a sound source coding unit for calculating a coding sound source from the input voice. 15. An LSP decoding means for decoding an LSP code and outputting a decoded LSP, and converting a frequency domain of the decoded LSP output from the LSP decoding means from a linear frequency domain to a frequency domain corresponding to auditory characteristics. Conversion means for converting; a correction means for correcting the decoded LSP whose frequency domain has been converted by the conversion means; an inverse conversion means for returning the frequency domain of the decoded LSP corrected by the correction means to the linear frequency domain; Comprising a sound source decoding means for decoding a sound signal to generate a sound source signal, and a synthesis means for generating a synthesized sound from the decoded LSP output from the inverse conversion means and the sound source signal generated by the sound source decoding means. Decryption device. 16. An LSP decoding unit for decoding an LSP code and outputting a decoded LSP, and a threshold value for a distance between adjacent dimensions is set based on each dimension value of the decoded LSP output from the LSP decoding unit. Threshold calculating means for calculating each dimension, correcting means for correcting the LSP based on the threshold for each dimension calculated by the threshold calculating means, and excitation decoding means for decoding an excitation code to generate an excitation signal A speech decoding device comprising: a decoding LSP corrected by the correction means; and a synthesis means for generating a synthesized sound from the sound source signal generated by the sound source decoding means. 17. An LSP decoding means for decoding an LSP code and outputting a decoded LSP, and converting a frequency domain of the decoded LSP output from the LSP decoding means from a linear frequency domain to a frequency domain corresponding to auditory characteristics. Conversion means for converting; a correction means for correcting the decoded LSP whose frequency domain has been converted by the conversion means; an inverse conversion means for returning the frequency domain of the decoded LSP corrected by the correction means to the linear frequency domain; Sound source decoding means for decoding the sound signal to generate a sound source signal; synthesizing means for generating a synthesized sound from the decoded LSP output from the LSP decoding means and the sound source signal generated by the sound source decoding means; Post decoding means for performing spectrum enhancement processing on the synthesized sound using the decoding LSP output from the conversion means. Device. 18. An LSP decoding unit for decoding an LSP code and outputting a decoded LSP, and a threshold value for a distance between adjacent dimensions is set based on each dimension value of the decoded LSP output from the LSP decoding unit. Threshold calculating means for calculating each dimension, correcting means for correcting the LSP based on the threshold for each dimension calculated by the threshold calculating means, and excitation decoding means for decoding an excitation code to generate an excitation signal Synthesizing means for generating a synthesized sound from the decoded LSP output from the LSP decoding means and the sound source signal generated by the sound source decoding means, and the synthesized sound using the decoded LSP corrected by the correcting means. And a post-filter means for performing a spectrum emphasis process on the audio signal.