JPH05346798A - Voice encoding device - Google Patents

Abstract

PURPOSE:To generate a driving sound code with which a pulse sound source with a period interval can be generated on the side of a decoding device without transmitting pitch period information. CONSTITUTION:An auditory weighting filter 11 generates a weighted input voice from an input voice A pitch extractor 12, on the other hand, extracts a pitch period from the input voice. A changeover switch 13 selects the pitch period or long-period predicted delay and outputs the selected information to an adaptive code book 14 and a pulse sound source generator 15. The adaptive code book 14 generates a long-period prediction signal and the pulse sound source generator 15 generates the pulse sound source. A driving sound source generator 16 generates a driving sound source with the long-period prediction signal and pulse sound source. A weighting synthesizing filter 17 generates a synthesized voice and a distortion minimizing unit 18 finds long-period predicted delay or pitch period which minimizes the sum of squares of an error of the synthesized voice based upon the weighted input voice and outputs the pulse position of the pulse sound source and the code of a gain.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding apparatus, and more particularly to a speech coding apparatus suitable for generating a driving excitation code from an input speech and using it for digital communication, voice mail and the like.

[0002]

2. Description of the Related Art In recent years, a CEL has been used as a voice encoding device at a low bit rate of about 4.8 to 8.0 kb / s.
P (Code-Excited Linear Pre)
A speech coding apparatus called a "diction coder" has been widely adopted. As a conventional CELP coding device of this type, as shown in FIG. 2, a device using a pulse sound source as a driving sound source is known.

In FIG. 2, reference numeral 21 is a perceptual weighting filter for perceptual weighting of input voice. Reference numeral 22 is a pitch extractor that extracts the pitch period from the input voice and outputs it. Reference numeral 23 is an applicable code book, which stores past driving sound sources. Reference numeral 24 is a pulse sound source generator that generates a pulse sound source. 25 is an adaptive codebook and pulse generator 2
4 is a driving sound source generator that generates a driving sound source from the output of FIG. Reference numeral 26 is a weighted synthesis filter for generating a weighted synthetic speech from the driving sound source. Reference numeral 27 is a distortion minimizer that calculates the error of the weighted synthetic speech with respect to the weighted input speech and outputs the sign of the long-term prediction delay, the pulse position, and the gain so as to minimize the sum of squares thereof.

The operation of the speech coding apparatus configured as above will be described below. First, the perceptual weighting filter 21 weights the input voice to obtain a weighted input voice. Next, the pitch extractor 22 extracts a pitch cycle from the input speech, transfers the pitch cycle to the decoding device side, and also the pulse excitation generator 24
Output to. The pulse sound source generator 24 generates a pulse sound source based on the pulse position and the pitch period. On the other hand, the adaptive codebook 23 generates a long-term prediction signal from the long-term prediction delay and the driving sound source generated in the past. The driving sound source generator 25 generates a driving sound source based on the long-term predicted signal and the pulse sound source. This driving sound source is weighted by the weighted synthesis filter 26 and converted into synthetic speech. This synthesized voice is the distortion minimizer 27.
, The distortion minimizer 27 calculates the sum of squared errors of the synthetic speech with respect to the weighted input speech.
And to minimize this calculated value, long-term prediction delay,
The pulse position and the gain of the driving sound source are determined respectively,
These codes are output as the driving excitation code. As a result, according to this apparatus, it is possible to generate the pitch period code and the driving excitation code from the input voice and output the information regarding these codes.

[0005]

However, in the conventional speech coding apparatus, the pitch period is used to generate the pulse excitation, so that the decoding apparatus side can use the pitch period code in addition to the driving excitation code. Information has to be transmitted, and there is a problem that the amount of transmitted information increases by that amount.

The present invention solves the above-mentioned conventional problems, and provides a speech coding apparatus capable of generating a pulse excitation having a periodic interval on the decoding apparatus side without transmitting pitch period information. The purpose is that.

[0007]

In order to achieve the above object, the present invention comprises a perceptual weighting means for performing perceptual weighting on an input voice to generate a weighted input voice signal,
Pitch cycle extraction means for extracting the pitch cycle of the input voice, selection means for selecting either long-term predicted delay information or pitch cycle information extracted by the pitch cycle extraction means in response to a command, and information selected by the selection means A long-term predicted signal generation means for generating a long-term predicted signal from the driving sound source information generated in the past, a pulse sound source generation means for generating a pulse sound source from information selected by the selection means and pulse position information, and a pulse sound source generation means The pulsed sound source by the generation and the long-term predicted signal generated by the long-term predicted signal generation means are each multiplied by a gain to generate a driving sound source, and a driving sound source generated by the driving sound source generation means is weighted. Synthesizing speech generating means for generating a weighted synthetic speech, and weighting synthesis for the weighted input speech signal A sum of squares calculating means for calculating a sum of squares of voice errors, a selection command means for instructing the selecting means to select one of a pitch period and a long-term prediction delay that minimizes the calculated value of the sum of squares calculating means, An information generating unit that generates information about a long-term predicted delay and a gain that minimizes the calculated value of the sum calculating unit, and a driving excitation code generating unit that generates a driving excitation code based on the information generated by the information generating unit are provided. This is a configuration of a speech encoding device that is installed.

[0008]

Therefore, according to the present invention, since the long-term prediction delay information for generating the long-term prediction signal is also used as the cycle interval information for generating the pulse sound source, the cycle interval information is transmitted to the decoding device side. It becomes unnecessary to transmit. further,
As the period interval information for generating the long-term predicted signal and the pulsed sound source, a selection means is provided to give a pitch period instead of the long-term predicted delay information. It is possible to generate a pulse train sound source having a more accurate cycle interval for different voice intervals.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows the configuration of an embodiment of the present invention. In FIG. 1, a speech coding apparatus includes a perceptual weighting filter 11, a pitch extractor 12, a changeover switch 13, an adaptive codebook 14, a pulse sound source generator 15, a driving sound source generator 16, a weighted synthesis filter 17, and a distortion minimizer. 1
8 is provided.

The perceptual weighting filter 11 is constructed as perceptual weighting means for performing perceptual weighting on the input voice to generate a weighted input voice signal. The pitch extractor 12 is configured as a pitch cycle extracting means for extracting a pitch cycle from the input voice. Changeover switch 13
Is configured as a selection unit that selects either the pitch period or the long-term prediction delay and outputs the selected information to the applicable codebook 14 and the pulse sound source generator 15.
The applicable codebook 14 is configured as a long-term predicted signal generating means for generating a long-term predicted signal from the information selected by the changeover switch 13 and the driving sound source information generated in the past. The pulse sound source generator 15 is configured as a pulse sound source generation unit that generates a pulse sound source based on the information selected by the changeover switch and the pulse position. The driving sound source generator 16 is configured as a driving sound source generating means that multiplies the long-term predicted signal and the pulse sound source by gains to generate a driving sound source from these signals. Weighted synthesis filter 17
Is configured as a synthetic voice generation means for weighting the driving sound source to generate a weighted synthetic voice. The distortion minimizer 18 constitutes a sum of squares calculating means for calculating the sum of squares of the error of the weighted synthesized speech with respect to the weighted input speech, and either the pitch period or the long-term prediction delay that minimizes the calculated value of the sum of squares. Selection command means for commanding one of the selections to the changeover switch 13 is configured. Further, information generating means for generating information about long-term predicted delay and gain that minimizes the calculated value of the sum of squares, and driving excitation code generating means for generating and outputting a driving excitation code based on the information generated by the information generating means. Is configured.

Next, the operation of the above embodiment will be described. First, when a voice is input to the voice encoding device, the weighted input voice is obtained by the perceptual weighting filter 11, and the pitch period is extracted from the input voice by the pitch extractor 12. Then, one of the pitch period and the long-term prediction delay is selected by the changeover switch 13, and the long-term prediction signal and the pulse sound source are generated based on the selected information. Then, the driving sound source is generated from the long-term predicted signal and the pulse sound source, the synthetic speech is generated based on the driving sound source, and the driving sound source code is generated based on the synthetic speech and the weighted input speech. When generating this drive excitation code, the drive excitation e (n) that generates the weighted synthesized speech closest to the weighted input speech is determined by distortion minimization, and encoding is performed according to this determination. Become.

Here, the driving sound source e (n) generated by the driving sound source generator 16 is the long-term predicted signal a (n-L) of the output of the applicable codebook 14, and the pulse sound source PM generated by the pulse sound source generator 15. It is composed of (n) and is represented by the following expression (1).

E (n) = β · a (n−L) + γ · PM (n) (1) where L is extracted by the long-term predicted delay 1 selected by the changeover switch 13 or the pitch extractor 12. Pitch period T
A period interval represented by p, PM (n) is a pulse train having impulses at a period interval L from the pulse position M, and β and γ are gains of the long-term predicted signal and the pulse sound source, respectively.

The procedure for determining the parameters L, M, β and γ of the driving sound source represented by the above equation (1) will be described below. First, the long-term prediction delay 1 is selected as the cycle interval L (L = 1, l = 1 min to lmax), and the long-term prediction signal a is selected.
For (n−L), the period interval L and the optimum gain β are determined by minimizing the distortion of the weighted synthetic speech. After these values are determined, the pulse position M and the optimum gain γ are determined by minimizing the distortion of the pulsed sound source PM (n) having the periodic interval L with respect to the remaining distortion.

Next, as the cycle interval L, the pitch cycle Tp
Is selected (L = Tp), and the optimum gain β of the long-term predicted signal at that time is determined. Then, after determining this, for the remaining distortion, the periodic interval L (= T
pulse position M of the pulsed sound source with p), optimal gain β
To decide. Then, finally, when the long-term predicted delay 1 or the pitch period Tp is selected as the cycle interval L, one of these pieces of information having a smaller distortion is selected and the selected code of L, β, M, γ is output. To do.

As described above, according to the above embodiment, since the long-term prediction delay l of the long-term predicted signal a (n-L) is used as the cycle interval information of the pulse sound source, the cycle interval information is transmitted to the decoding device side. It becomes unnecessary to transmit. Furthermore, since the changeover switch 13 that can give the pitch period Tp instead of the long-term predicted delay 1 as the long-term predicted delay 1 of the long-term predicted signal and the period interval information L of the pulse sound source is provided, there is no sound (unvoiced sound). It is possible to generate a pulse train sound source having a more accurate cycle interval for a voice section such as a rising portion to a voice sound.

[0017]

As is apparent from the above embodiment, the present invention uses the long-term prediction delay information for generating the long-term prediction signal as the period interval information for generating the pulse sound source, and therefore the decoding is performed. It is not necessary to transmit the cycle interval information to the device side. Further, as the long-term prediction delay information for generating the long-term prediction signal and the cycle interval information for generating the pulse sound source, a selecting means for selecting the pitch period is provided instead of the long-term prediction delay information, so that there is no sound or unvoiced sound. For the voice section such as the rising part to the voice sound,
It is possible to generate a pulse sound source having a more accurate cycle interval.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of a speech encoding apparatus according to an embodiment of the present invention.

FIG. 2 is a schematic block diagram of a conventional speech encoding device.

[Explanation of symbols]

 11 Perceptual Weighting Filter 12 Pitch Extractor 13 Changeover Switch 14 Applicable Codebook 15 Pulse Sound Source Generator 16 Drive Sound Source Generator 17 Weighted Synthesis Filter 18 Distortion Minimizer

Claims (1)

[Claims] 1. A perceptual weighting means for performing perceptual weighting on an input voice to generate a weighted input voice signal,
Pitch cycle extraction means for extracting the pitch cycle of the input voice, selection means for selecting either long-term predicted delay information or pitch cycle information extracted by the pitch cycle extraction means in response to a command, and information selected by the selection means A long-term predicted signal generation means for generating a long-term predicted signal from the driving sound source information generated in the past, a pulse sound source generation means for generating a pulse sound source from information selected by the selection means and pulse position information, and a pulse sound source generation means The pulsed sound source by the generation and the long-term predicted signal generated by the long-term predicted signal generation means are each multiplied by a gain to generate a driving sound source, and a driving sound source generated by the driving sound source generation means is weighted. Synthesizing speech generating means for generating a weighted synthetic speech, and weighting synthesis for the weighted input speech signal A sum of squares calculating means for calculating a sum of squares of voice errors, a selection command means for instructing the selecting means to select one of a pitch period and a long-term prediction delay that minimizes the calculated value of the sum of squares calculating means, An information generating unit that generates information about a long-term predicted delay and a gain that minimizes the calculated value of the sum calculating unit, and a driving excitation code generating unit that generates a driving excitation code based on the information generated by the information generating unit are provided. Speech coding device.