JP2002149199A - Unit for generating sound source - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the deterioration of voice quality owing to the attenuation of signal power which relies on a frequency band to be conspicuous especially when a wide band voice signal is inputted in a sound source generating unit which is required in a digital telephone. SOLUTION: In an ACELP voice encoding system sound source generating unit, a pulse vector generating part 13 generates a pulse string in accordance with a algebraic rule and a pulse vector extending part 14 performs the convolution of extension patterns stored in an extension pattern storing part 12 concerning the pulse string. Then the distribution of pulse string energy is diffused in terms of time and also a desired frequency characteristic is given. The extension pattern storing part 12 stores the extension patterns provided with a high- pass characteristic. By the configuration, the energy distribution of a sound source signal on a frequency axis is diffused to a whole frequency band so as to realize the sound source generating device for improving the quality of decoding voice.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a CELP (Code Excited LTE) used for digital mobile communication and voice storage.
The present invention relates to a sound source generation device in a speech codec.

[0002]

2. Description of the Related Art In the field of digital mobile communication and voice storage, a voice coding apparatus for compressing voice information for efficient use of radio waves and storage media and coding the voice information with high efficiency is used. Above all, a system based on the CELP system has been widely put into practical use at medium and low bit rates. In addition, about the technology of CELP, for example, MRSc
hroeder and bsAtal: "Code-Excite Linear Predictio
n (CELP): High-quality Speech at Very Low Bit Rate
s ", Proc. ICASSP-85, 25.1.1, pp. 937-940, 1985".

[0003] In the CELP type speech coding system, a digitized speech signal is converted to a fixed frame length (5 m to 50 m).
s), the speech is linearly predicted for each frame, and the prediction residual (excitation signal) by the prediction is encoded using an adaptive codebook having a known waveform and a noise codebook. The adaptive codebook stores excitation signals generated in the past, and is used to represent periodic components of a speech signal.

FIG. 4 is a block diagram showing a configuration of a sound source generating apparatus in a conventional CELP speech codec.
As shown in FIG. 1, a conventional excitation generator includes an adaptive codebook unit 1, an adaptive codebook selection unit 2, an adaptive gain adjustment unit 3
, A random codebook section 4, a random codebook selection section 5, a noise gain adjustment section 6, an addition section 7, and an adaptive codebook update section 8.

The adaptive codebook unit 1 has a storage device such as a semiconductor memory and stores excitation vectors. Adaptive codebook selecting section 2 selects one excitation vector from adaptive codebook section 1. Adaptive gain adjuster 3 multiplies the excitation vector selected by adaptive codebook selector 2 by a constant and outputs the result. The noise codebook unit 4 accumulates excitation vectors determined by learning. The random codebook selecting unit 5 selects one excitation vector from the random codebook unit 4. The noise gain adjustment unit 6 multiplies the excitation vector selected by the noise codebook unit 4 by a constant. Adder 7
Generates an excitation signal by adding the adaptive excitation vector whose gain has been adjusted by the adaptive gain adjustment unit 3 and the noise excitation vector whose gain has been adjusted by the noise gain adjustment unit 6. Adaptive codebook updating section 8 updates the excitation vector of adaptive codebook section 1 with the excitation signal generated by adding section 7.

Next, the operation of the sound source generating apparatus having the above configuration will be described. In an analysis by synthesis (AbS: Analysis by Synthesis) type speech coding method represented by CELP, a decoded speech signal that can be generated by driving speech synthesis model parameters in various combinations is used.
By encoding a parameter value group that minimizes an error from the input audio signal, a high compression rate and high decoded audio quality, that is, high-efficiency audio encoding is realized.

[0007] The signal source of the sound source generator in the CELP speech coding system is composed of the sum of two types of signal sources, a signal source for generating a new signal and a signal source for restoring a signal generated in the past.

First, an excitation vector is selected by adaptive codebook selecting section 2 from adaptive codebook section 1 as a signal source based on a past signal. The selected sound source vector is multiplied by a constant in the adaptive gain adjustment unit 3 and output as an adaptive vector. On the other hand, an excitation vector is selected by a random codebook selecting unit 5 from a predetermined random codebook unit 4 as a signal source for generating a new signal. The selected sound source vector is multiplied by a constant in the noise gain adjustment unit 6 and output as a noise vector. These two vectors are added by the adder 7 and output as a sound source signal. Here, adaptive codebook section 1 accumulates the excitation signal output in the past, and is updated by adaptive codebook updating section 8 every time the excitation signal used for speech decoding is determined.

As described above, by updating the codebook using the signal generated in the past, it becomes possible to improve the coding efficiency for a steady signal. With the above operation, it is possible to generate a sound source signal with a small quantization error even at a low bit rate.

[0010]

However, in the conventional sound source generating apparatus, no design or encoding selection is made in consideration of the frequency characteristics of the generated sound source.
In order to design or select such that the temporal signal-to-noise ratio, i.e., SN ratio, of the input audio signal and the decoded audio signal is minimized,
There is a problem in that a codebook is designed or selected so as to minimize an error with respect to a low-frequency band signal having a large power, and that the SN ratio for a high-frequency band signal is greatly deteriorated. In particular, this problem becomes remarkable in voice codec processing for inputting a wideband voice signal having a signal band of 50 Hz to 7 kHz.

The present invention has been made in view of the above points, and a sound source generating apparatus which achieves a desired SN ratio for each frequency band by designing or selecting a codebook in consideration of frequency characteristics. The purpose is to realize.

[0012]

According to a first aspect of the present invention, there is provided an excitation generating apparatus capable of updating an excitation vector with an excitation signal generated in the past, and selecting one excitation vector from the adaptive code storage. Adaptive codebook selecting means, an adaptive vector filter for filtering the excitation vector selected by the adaptive codebook selecting means so as to have a desired frequency characteristic, and adaptation for multiplying the excitation vector filtered by the adaptive vector filter by a constant. Gain adjustment means, a random code storage means for storing the excitation vector obtained in advance by learning, a random codebook selection means for selecting one excitation vector from the random code storage means, and a random codebook selection means A noise vector that filters a selected sound source vector to have a desired frequency characteristic. A noise filter, a noise gain adjusting means for multiplying the sound source vector filtered by the noise vector filter by a constant, and a sound source vector gain-adjusted by the adaptive gain adjusting means and a sound source vector gain-adjusted by the noise gain adjusting means. The present invention employs a configuration including an adding unit that generates an excitation signal by adding, and an adaptive codebook updating unit that updates an adaptive excitation vector of the adaptive code storage unit with the excitation signal generated by the adding unit.

According to this configuration, there are provided an adaptive vector filter for filtering an adaptive excitation vector so as to have a desired frequency characteristic, and a noise vector filter for filtering a noise excitation vector so as to have a desired frequency characteristic. Therefore, the frequency characteristics of the generated sound source can be manipulated. This makes it possible to realize a sound source generation device that improves the quality of decoded speech.

According to another aspect of the present invention, there is provided an adaptive code storage means capable of updating an adaptive excitation vector with a previously generated excitation signal, and an adaptive code selecting one adaptive excitation vector from the adaptive code storage means. Book selection means, adaptive gain adjustment means for multiplying the adaptive excitation vector selected by the adaptive codebook selection means by a constant, noise code storage means storing a vector in which energy is concentrated in a high frequency band, and the noise A noise codebook selecting means for selecting one noise excitation vector from the code accumulating means, a noise gain adjusting means for multiplying the noise excitation vector selected by the noise codebook selecting means by a constant, and a gain adjustment by the adaptive gain adjusting means. Adding means for adding the adjusted adaptive excitation vector and the noise excitation vector whose gain has been adjusted by the noise gain adjusting means to generate an excitation signal. When, a configuration having a an adaptive codebook updating means for updating the adaptive code storage means by means of the sound source signal generated by the adding means.

According to this configuration, since the noise codebook which stores the vector in which the energy is concentrated in the high frequency band signal is provided, the high frequency band signal can be obtained without increasing the processing amount as compared with the conventional sound source generating apparatus. Band signal attenuation can be suppressed. This makes it possible to realize a sound source generation device that improves the quality of decoded speech.

[0016] The recording medium of the present invention includes a computer comprising: an adaptive codebook selecting means, an adaptive gain adjusting means, a noise codebook selecting means, a noise gain adjusting means, an adding means, A program for functioning as book updating means is recorded.

According to this configuration, it is possible to cause a conventional sound source generating apparatus that performs software processing using a computer to function as the sound source generating apparatus of the present invention. Thus, a sound source generation device can be realized at a minimum cost.

An excitation generator according to the present invention comprises an adaptive code storage means capable of updating an excitation vector with an excitation signal generated in the past, and an adaptive codebook selection means for selecting one excitation vector from the adaptive code storage means. An adaptive gain adjusting means for multiplying the excitation vector selected by the adaptive codebook selecting means by a constant, a spreading pattern accumulating means for accumulating a vector in which energy is concentrated in a high frequency band, and generating a vector by a plurality of pulses. A pulse vector generating unit, a pulse vector spreading unit that convolves the pulse vector generated by the pulse vector generating unit with a vector stored in the diffusion pattern storage unit, and a pulse vector that is spread by the pulse vector spreading unit. Noise gain adjusting means for multiplying by a constant, and gain adjusted by the adaptive gain adjusting means. Adding means for adding a source vector and a pulse vector whose gain has been adjusted by the noise gain adjusting means to generate an excitation signal; and adaptive codebook updating means for updating the adaptive code storage means with the excitation signal generated by the adding means. Are adopted.

According to this configuration, since the spread pattern storing means for storing the vector in which the energy is concentrated in the high frequency band signal is provided, the ACLP (Algebraic CELP) speech coding system having a small storage capacity of the codebook is provided. Also, the attenuation of the high frequency band signal can be suppressed. This makes it possible to realize a sound source generation device that improves the quality of decoded speech.

[0020] The recording medium of the present invention comprises a computer comprising: an adaptive codebook selecting means, an adaptive gain adjusting means, a pulse vector generating means, a pulse vector spreading means, a noise gain adjusting means, Means and a program for functioning as an adaptive codebook updating means are recorded.

According to this configuration, it is possible to cause a conventional sound source generating apparatus that performs software processing using a computer to function as the sound source generating apparatus of the present invention. Thus, the sound source generation device of the present invention can be realized at a minimum cost.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The gist of the present invention is that the adaptive excitation vector and the noise excitation vector are each filtered so as to obtain desired frequency characteristics, and the attenuation of a high frequency band signal is suppressed. The purpose is to prevent dependent power loss and improve the quality of decoded speech.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

(Embodiment 1) FIG. 1 is a block diagram showing a configuration of a sound source generating apparatus according to Embodiment 1 of the present invention. In this figure, the same parts as those in FIG. 4 described above are denoted by the same reference numerals.

In this figure, an excitation generator according to the present embodiment includes an adaptive codebook section (adaptive code storage means) 1, an adaptive codebook selecting section 2, an adaptive vector filter 9, an adaptive gain adjusting section 3, Noise code book section (noise code storage means) 4
, A noise codebook selector 5, a noise vector filter 10
, A noise gain adjusting unit 6, an adding unit 7, and an adaptive codebook updating unit 8.

The adaptive codebook unit 1 has a storage device such as a semiconductor memory and stores excitation vectors. Adaptive codebook selecting section 2 selects one adaptive excitation vector from adaptive codebook section 1. The adaptive vector filter 9 filters the adaptive excitation vector so as to have a desired frequency characteristic. The adaptive gain adjuster 3 multiplies the filtered adaptive excitation vector by a constant. The noise codebook unit 4 stores the excitation vectors obtained in advance by learning.

The random codebook selecting section 5 selects one noise excitation vector from the random codebook section 1. The noise vector filter 10 filters the noise source vector so as to have a desired frequency characteristic. The noise gain adjuster 6 multiplies the filtered noise source vector by a constant. The adding unit 7 generates and outputs a sound source signal by adding the adaptive sound source vector and the noise sound source vector, each of which has been adjusted in gain. Adaptive codebook updating section 8 updates the excitation vector of adaptive codebook section 1 with the excitation signal generated by adding section 7.

Next, the operation of the sound source generating apparatus having the above configuration will be described. First, an excitation vector is selected by adaptive codebook selecting section 2 from adaptive codebook section 1 as a signal source based on a past signal. The selected sound source vector is shaped by a filter having a predetermined frequency characteristic in the adaptive vector filter 9, further multiplied by a constant in the adaptive gain adjustment unit 3, and output as a sound source vector generated by the adaptive sound source.

On the other hand, an excitation vector is selected by the random codebook selecting unit 5 from the random codebook unit 4 as a signal source for generating a new signal. The selected sound source vector is shaped by a filter having a predetermined frequency characteristic by a noise vector filter 10, further multiplied by a constant by a noise gain adjustment unit 6, and output as a sound source vector generated by a noise sound source. These two vectors are added by the adder 7 and output as a sound source signal.

Here, adaptive codebook section 1 accumulates the excitation signal output in the past, and is updated by adaptive codebook updating section 8 every time the excitation signal used for speech decoding is determined.

With the above operation, it is possible to control the frequency characteristics of the generated sound source, and it is possible to realize a sound source generating apparatus which can improve the quality of decoded speech.

In the above embodiment, at least adaptive codebook selector 2, adaptive gain adjuster 3, noise codebook selector 5, noise gain adjuster 6, and adder 7
And a conventional sound source that performs software processing using a computer by programming the function of the adaptive codebook updating unit 8 and recording it on a recording medium such as a semiconductor memory, a magnetic storage medium, an optical recording medium, or a magneto-optical recording medium. The generation device can function as the sound source generation device of the present embodiment.

(Embodiment 2) FIG. 2 is a block diagram showing a configuration of a sound source generating apparatus according to Embodiment 2 of the present invention. In this figure, parts common to FIG. 4 described above are denoted by the same reference numerals and description thereof is omitted.

The sound source generating apparatus according to the present embodiment has the same configuration as the conventional sound source generating apparatus shown in FIG. 4, but has a noise codebook section 11 storing a vector in which energy is concentrated in a high frequency band signal. In that it has With the noise codebook section 11, it is possible to suppress the attenuation of the high frequency band signal without increasing the processing amount as compared with the conventional sound source generation device.

(Embodiment 3) FIG. 3 is a block diagram showing a configuration of a sound source generating apparatus according to Embodiment 3 of the present invention. In this figure, parts common to FIG. 4 described above are denoted by the same reference numerals and description thereof is omitted.

In this figure, the excitation generating apparatus according to the present embodiment includes an adaptive codebook unit 1, an adaptive codebook selecting unit 2, an adaptive gain adjusting unit 3, a spreading pattern accumulating unit 12, a pulse vector generating unit. 13, a pulse vector spreading unit 14,
It comprises a noise gain adjuster 6, an adder 7, and an adaptive codebook updater 8.

The spreading pattern storage unit 12 stores a vector in which energy is concentrated in the high frequency band signal. The pulse vector generation unit 13 generates a vector using a plurality of pulses. The pulse vector spreading unit 14 convolves the pulse vector generated by the pulse vector generation unit 13 with the vector stored in the diffusion pattern storage unit 12. The noise gain adjuster 6 multiplies the spread pulse vector by a constant.
The adder 7 generates and outputs a sound source signal by adding the adaptive sound source vector and the pulse vector, each of which has been adjusted in gain.

Next, the operation of the sound source generating apparatus having the above configuration will be described. The adaptive codebook unit 1 stores the excitation signal generated in the past using the sample point as an index. An adaptive vector is selected by the adaptive codebook selecting unit 2 from the adaptive codebook unit 1. The adaptive vector selected in adaptive codebook section 2 is multiplied by a constant in adaptive gain adjustment section 3 and output as an excitation vector. Since adaptive codebook section 1 stores the excitation signal output in the past, it is updated by adaptive codebook updating section 8 every time the excitation signal used for speech decoding is determined.

In the CELP speech coding method, a sound source signal is selected on the basis of minimizing an error on the time axis between an input signal and a synthesized signal. The sound source signal that minimizes the band error is selected. Therefore, the vector stored in adaptive codebook section 1 has much energy in the low frequency band.

On the other hand, in the pulse vector generator 13, for example, in ACELP, a pulse train having an amplitude of "1" or "-1" is generated according to an algebraic rule. By convolving the diffusion pattern stored in the diffusion pattern storage unit 12 with the pulse vector spreading unit 14 with respect to this pulse train,
It is possible to temporally disperse the energy distribution of the pulse train and provide a desired frequency characteristic. As described above, since the adaptive sound source vector is predicted to have large energy in a low frequency band, a diffusion pattern having a high-pass characteristic is stored in the diffusion pattern storage unit 12.

This diffusion pattern is predetermined.
Learning can be performed by obtaining a diffusion pattern so that the SN ratio for each frequency band is constant and minimum for predetermined data. The spread pulse train is multiplied by a constant in the noise gain adjusting unit 6 and output as a sound source vector. Then, these two vectors are added by the adder 7 and output as a sound source signal.

With the above operation, even in the ACLP (Algebraic CELP) speech coding system having a small storage capacity of the adaptive codebook, the attenuation of the high frequency band signal can be suppressed without increasing the processing amount. This makes it possible to realize a sound source generation device that improves the quality of decoded speech.

In the above embodiment, at least adaptive codebook selecting section 2, adaptive gain adjusting section 3, pulse vector generating section 13, pulse vector spreading section 14
The functions of the noise gain adjusting unit 6, the adding unit 7, and the adaptive codebook updating unit 8 are programmed and recorded on a recording medium such as a semiconductor memory, a magnetic storage medium, an optical recording medium, or a magneto-optical recording medium. A conventional sound source generating apparatus that performs software processing using a computer can function as the sound source generating apparatus according to the present embodiment.

[0044]

As described above, according to the present invention,
It is possible to suppress the attenuation of the high frequency band signal, which has occurred in the conventional AbS type speech codec processing, and to improve the quality of the decoded speech.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a sound source generation device according to Embodiment 1 of the present invention.

FIG. 2 is a block diagram showing a configuration of a sound source generation device according to Embodiment 2 of the present invention.

FIG. 3 is a block diagram showing a configuration of a sound source generation device according to Embodiment 3 of the present invention.

FIG. 4 is a block diagram showing a configuration of a sound source generation device in a conventional CELP speech codec;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Adaptive codebook part 2 Adaptive codebook selection part 3 Adaptive gain adjustment part 4, 11 Noise codebook part 5 Noise codebook selection part 6 Noise gain adjustment part 7 Addition part 8 Adaptive codebook update part 9 Adaptive vector filter 10 Noise vector Filter 12 Diffusion pattern storage unit 13 Pulse vector generation unit 14 Pulse vector diffusion unit

Claims (5)

[Claims] 1. An adaptive code storage means capable of updating an excitation vector with an excitation signal generated in the past, an adaptive codebook selection means for selecting one excitation vector from the adaptive code storage means, and an adaptive codebook selection means. An adaptive vector filter for filtering the sound source vector selected by the means so as to have a desired frequency characteristic, an adaptive gain adjusting means for multiplying the sound source vector filtered by the adaptive vector filter by a constant, and a sound source vector previously obtained by learning. , A noise codebook selecting means for selecting one excitation vector from the noise code storage means, and an excitation vector selected by the noise codebook selection means having a desired frequency characteristic. Vector filter for filtering as described above, and the noise vector filter Noise gain adjusting means for multiplying a sound source vector filtered by a constant, generating a sound source signal by adding the sound source vector gain-adjusted by the adaptive gain adjusting means and the sound source vector gain-adjusted by the noise gain adjusting means. A sound source generating apparatus comprising: an adding means for performing adaptive codebook updating; and an adaptive codebook updating means for updating an adaptive excitation vector of the adaptive code accumulating means with the excitation signal generated by the adding means. 2. An adaptive code storage means capable of updating an adaptive excitation vector by an excitation signal generated in the past; an adaptive codebook selecting means for selecting one adaptive excitation vector from the adaptive code storage means; An adaptive gain adjusting means for multiplying the adaptive excitation vector selected by the book selecting means by a constant, a noise code storage means storing a vector in which energy is concentrated in a high frequency band, and a noise excitation vector from the noise code storage means. A noise codebook selecting means for selecting one, a noise gain adjusting means for multiplying the noise excitation vector selected by the noise codebook selecting means by a constant, an adaptive excitation vector gain-adjusted by the adaptive gain adjusting means, and the noise Adding means for generating a sound source signal by adding the noise sound source vector gain-adjusted by the gain adjusting means; and An adaptive codebook updating means for updating the adaptive code storage means with an excitation signal. 3. The computer according to claim 1 or 2,
A program that functions as an adaptive codebook selecting means, an adaptive gain adjusting means, a noise codebook selecting means, a noise gain adjusting means, an adding means, and an adaptive codebook updating means of the sound source generating apparatus described in the above. Recording medium characterized by the above-mentioned. 4. An adaptive code storage means capable of updating an excitation vector with an excitation signal generated in the past, an adaptive codebook selection means for selecting one excitation vector from the adaptive code storage means, and an adaptive codebook selection means. Adaptive gain adjusting means for multiplying the sound source vector selected by the means by a constant, diffusion pattern accumulating means for accumulating a vector in which energy is concentrated in a high frequency band, and pulse vector generating means for generating a vector by a plurality of pulses. A pulse vector spreading means for convolving the pulse vector generated by the pulse vector generating means with a vector stored in the diffusion pattern storing means; and a noise gain adjustment for multiplying the pulse vector spread by the pulse vector spreading means by a constant. Means, a sound source vector whose gain has been adjusted by the adaptive gain adjustment means, and the noise source vector. An adding means for adding the pulse vector gain-adjusted by the in-adjustment means to generate an excitation signal; and an adaptive codebook updating means for updating the adaptive code storage means with the excitation signal generated by the adding means. A sound source generator characterized by the above-mentioned. 5. The computer according to claim 3, further comprising: an adaptive codebook selecting unit, an adaptive gain adjusting unit, a pulse vector generating unit, a pulse vector spreading unit, a noise gain adjusting unit, and an adding unit. And a program that functions as an adaptive codebook updating means.