JPH08248999A - Voice coding/decoding device - Google Patents

Abstract

PURPOSE: To generate a decoded voice source signal with excellent continuity and to obtain the decoded voice source signal with high quality by providing a voice source shaping part weighting and adding a representative voice source of a decoded frame and an impulse signal. CONSTITUTION: The voice source shaping part 25 being a voice source shaping means is provided in a decoding part 120, and the impulse signal is weighted and added to the representative voice source S20, and the representative voice source S20 is shaped. Since the waveform shape of the shaped representative voice source S26 shaped by the shaping approaches to an impulse always, even when that is repeated in a representative voice source repeat part 21, the decoded voice source signal S22 with a smooth change is generated. Since the representative voice source S20 is shaped so as to approach to the impulse by the voice source shaping part 25 in such a manner, the decoded voice source signal with the smooth change between frames is obtained. Further, since the shaping is performed so as to approach to the impulse further when a change of an input voice signal is large, the decoded voice source signal following up even large change is obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding / decoding apparatus used for converting a speech signal into a digital signal for transmission or storage.

[0002]

2. Description of the Related Art Conventionally, a voice signal is analyzed for each frame of a predetermined length, separated into a sound source signal and spectrum envelope information and encoded, and the encoded data is decoded to generate a voice signal. As a speech coding / decoding device, for example, Document 1 (Takahashi, Iimori "Vocoder Using Vector Quantization of Phase Equalized Sound Source", Japan Electrical Association Hokuriku Branch Union Conference, B145, P291, October 1992) The one shown in is known.

This conventional example takes advantage of the fact that a similar waveform is repeated in a pitch cycle when a voice signal is a voiced sound, and the sound source signal of the frame is represented by only a signal of one pitch cycle length in it. This is characterized by improving the coding characteristics of the voiced sound part.By equalizing the phase of the excitation signal for a short time, the extraction of the representative excitation is stabilized and the vector quantization of the representative excitation is made more efficient. There is an advantage that it can be achieved.

FIG. 4 is a block diagram showing the configuration of the above-mentioned conventional speech coding / decoding apparatus. As shown in the figure,
The audio encoding / decoding device 200 is configured to include an encoding unit 210 and a decoding unit 220. The coding unit 210 described above includes a spectrum envelope analysis unit 2, an inverse filter unit 4, a phase equalization unit 7, a representative sound source cutout unit 9, a pitch period extraction unit 11, a voiced / unvoiced determination unit 13, and It has a representative excitation encoding unit 17. In addition, the decoding unit 22 described above
0 represents the representative excitation decoding unit 19 and the representative excitation repetition unit 21.
And a synthesis filter unit 23.

Next, the operation of the above-described conventional speech coding / decoding apparatus 200 will be described. First, the encoding unit 210
The operation of will be described. The spectrum envelope analysis unit 2
The input voice signal S1 is analyzed for each input frame, and the spectrum envelope of the voice signal of the corresponding frame is calculated as
For example, it is obtained by a linear predictive analysis method and is output to the inverse filter unit 4 as spectrum envelope information S3. At the same time, the spectrum envelope analysis unit 2 encodes this spectrum envelope information S3 and outputs it to the outside as a spectrum envelope code S5.

The voiced / unvoiced decision unit 13 receives the input voice signal S
For example, the power of 1 is analyzed to determine whether the input voice signal S1 is voiced or unvoiced, and the determination result is output to the pitch period extraction unit 11 as voiced / unvoiced determination information S12. At the same time, the voiced / unvoiced determination unit 13 encodes this voiced / unvoiced determination information S12 and outputs it as a voiced / unvoiced determination code S16 to the outside. When the voiced / unvoiced determination information S12 is voiced sound, the pitch cycle extraction unit 11 performs a pitch cycle analysis on the input voice signal S1 and extracts the obtained pitch cycle S10 from the representative sound source cutout unit 9
Output to. In addition, the pitch period extraction unit 11 simultaneously
The pitch cycle S10 is encoded and output as a pitch cycle code S15 to the outside.

The inverse filter unit 4 uses the above-mentioned spectrum envelope information S3 to perform, for example, linear prediction inverse filter processing on the input voice signal S1 to obtain a sound source signal S6, and outputs it to the phase equalization unit 7. The phase equalizer 7 performs a short-time phase equalization process so that the phase of the sound source signal S6 becomes zero at the maximum amplitude position (peak position) on the waveform appearing at pitch period intervals, and the phase equalized sound source signal S8 is obtained. Is output to the representative sound source cutout unit 9. The representative sound source cutout unit 9 has a pitch cycle S.
When 10 is input, that is, voiced / unvoiced determination information S12
Is a voiced sound, a signal corresponding to the length of one pitch period S10 is cut out from the phase-equalized sound source signal S8 with reference to the maximum amplitude position of the phase-equalized sound source signal 8 in the corresponding frame. Representative excitation encoding unit 17 as excitation S14
Output to. The representative excitation encoding unit 17 vector-quantizes and encodes the above-described representative excitation S14, and outputs the obtained representative excitation code S18 to the outside.

Next, the operation of the decoding section will be described.
The representative excitation decoding unit 19 encodes the representative excitation code S
18 is decoded, and the decoded representative sound source S20 is used as the representative sound source repeating unit 2
Output to 1. The representative sound source repeating unit 21 decodes the voiced / unvoiced determination code S16 and the pitch period code S15 into voiced / unvoiced determination information and pitch period, respectively, and when the voiced / unvoiced determination information is voiced sound, the pitch period The decoded representative sound source S20 is repeated at intervals of to generate a decoded sound source signal S27,
Output to the synthesis filter 23. When the voiced / unvoiced determination information is unvoiced sound, white noise is generated and output to the synthesis filter 23 as a decoded excitation signal S27. The synthesis filter unit 23 decodes the spectrum envelope from the spectrum envelope code S5 and synthesizes the decoded speech signal S28 from the decoded spectrum envelope and the decoded excitation signal S27 by, for example, a linear prediction synthesis filter method.

[0009]

However, in the conventional speech coding / decoding apparatus 200 as shown in FIG. 4, the representative excitation repeating unit 21 of the decoding unit 220 simply outputs the decoded representative excitation S20 to the pitch cycle. Since the decoded excitation signal S27 of the corresponding frame is repeatedly generated at intervals, if the shapes of the decoded representative excitations of the corresponding frame and the frames immediately before and after the same differ to some extent or more, the decoded excitation signal suddenly changes at the frame boundary. As a result, there is a problem that sound quality deterioration (abnormal noise due to abrupt change) occurs in the decoded speech signal.

This will be described with reference to FIG. Figure 5
5A and 5B are diagrams for explaining generation of a decoded excitation signal by simple repetition of a decoded representative excitation in the above conventional speech encoding / decoding device. FIG. 5A shows a representative excitation and FIG. 5B shows a decoded excitation signal. Is shown. As shown in the figure, when the shapes of the decoded representative sound sources of the corresponding frame M and the immediately preceding and succeeding frames (M-1) and (M + 1) are different, frame boundaries B3 and B3
At 4, the decoded sound source signal suddenly changes, resulting in deterioration of the sound quality of the decoded speech signal (such as abnormal noise due to a sudden change).

The present invention has been made to solve the above problems, and generates a decoded excitation signal with good continuity even if the decoding representative excitation of the corresponding frame and the immediately preceding and succeeding frames are different, and It is an object of the present invention to provide a speech coding / decoding device capable of obtaining a good decoded speech signal.

[0012]

In order to solve the above-mentioned problems, the present invention according to claim 1 analyzes an input voice signal for each frame of a predetermined length, and obtains a spectrum envelope information and a sound source of the input voice signal. A signal and a pitch period are obtained, and when the input voice signal is a voiced sound, short-time phase equalization is performed so that the phase of the sound source signal becomes zero at the maximum amplitude position on the signal waveform appearing at the intervals of the pitch period. An encoding unit that extracts a representative sound source for one pitch period length after performing the processing, encodes the representative sound source, the pitch period, and the spectrum envelope information, and outputs the spectrum envelope code.
Each encoded parameter output from this encoding unit is decoded, and when the input speech signal is a voiced sound, the decoded representative sound sources of one pitch cycle length are arranged at intervals of the pitch cycle. And a decoding unit that decodes the decoded speech signal by using this decoded excitation signal and the decoded spectrum envelope, and is decoded by the speech coding and decoding apparatus. Sound source shaping means for weighting and adding the representative sound source and impulse signal of the corresponding frame, and a representative sound source repetition for generating the decoded sound source signal by arranging the shaped representative sound sources output from the sound source shaping means at intervals of the pitch cycle. And a means for providing the decoding unit.

According to a second aspect of the present invention, in the voice encoding / decoding apparatus according to the first aspect, the excitation shaping means is configured to change the voice signal of the corresponding frame and a frame in the vicinity of the corresponding frame. It is configured to change the value of the weighting coefficient when performing weighted addition of the representative sound source and the impulse signal according to the magnitude.

[0014]

According to the invention having the above-mentioned structure, the sound source shaping means weights and adds the representative sound source and the impulse signal of the decoded corresponding frame. Further, the representative excitation repeating unit generates the decoded excitation signal by arranging the shaped representative excitations output from the excitation shaping unit at intervals of the pitch cycle.

According to the invention of claim 2 having the above-mentioned configuration, the sound source shaping means determines the representative sound source and the representative sound source according to the magnitude of change in the audio signal of the corresponding frame and a frame in the vicinity of the corresponding frame. The value of the weighting coefficient when weighting and adding the impulse signal is changed.

[0016]

【Example】

(First Embodiment) An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a speech coding / decoding apparatus that is a first embodiment of the present invention. As shown in the figure, this audio encoding / decoding apparatus 101 is configured to include an encoding unit 110 and a decoding unit 120. The encoding unit 110 includes the spectrum envelope analysis unit 2, the inverse filter unit 4, the phase equalization unit 7, and the representative sound source cutout unit 9.
A pitch period extraction section 11 and a voiced / unvoiced determination section 13
And a representative excitation coding unit 17. Further, the above decoding unit 120 includes the representative excitation decoding unit 19, the representative excitation repeating unit 21, the excitation shaping unit 25, and the synthesis filter unit 2.
It is configured to have 3. The configuration of the speech coding / decoding apparatus 101 of the first embodiment differs from that of the speech coding / decoding apparatus 200 of the conventional example described above in that the decoding unit 1
20 is provided with a sound source shaping section 25 which is a sound source shaping means. This excitation shaping section 25 is arranged on the output side of the representative excitation decoding section 19 and on the input side of the representative excitation repetition section 21, which is the representative excitation repetition means. The components other than the sound source shaping unit 25 have the same configurations and the same effects as the components shown in FIG. 4, and therefore, the same reference numerals as those of the components shown in FIG.

Hereinafter, the above speech coding / decoding device 101 will be described.
The operation of will be described. The sound source shaping unit 25 shapes the representative sound source S20 by weighting and adding the impulse signal to the representative sound source S20. By this shaping, the shaped waveform of the shaped representative sound source S26 is always close to an impulse,
Even if the representative excitation repeating unit 21 repeats this, a decoded excitation signal S22 with a smooth change can be generated. Also, when the input audio signal changes significantly between the current frame and its neighboring frames, the above-mentioned weighting coefficient is adjusted to bring the representative sound source closer to the impulse, thereby smoothing the change of the sound source signal. You can

The operation of the sound source shaping unit 25 will be described below.
This will be described in more detail. Here, the representative sound source sound S20 is RC
i (i is a positive integer from 1 to P), the impulse signal having a pulse at the same position as the maximum amplitude position of the representative sound source S20 is RIi, the pitch period is P, and the weighting coefficient is α (α is 0 ≦ α ≦ 1), the sound source shaping unit 25 weights and adds both according to the following formula RMi = (1-α) × RCi + α × RIi ... (1) and the shaped representative sound source S26 Obtain a signal RMi.

In the above equation (1), the following equation

[Equation 1] There is a relationship.

FIG. 2 is a diagram for explaining the operation of weighted addition of the representative sound source expressed by the above equation (1) in the sound source shaping section 25. As shown in the figure, it is shown that a shaped representative sound source that approaches the shape of the impulse waveform can be obtained.

Further, the sound source shaping unit 25 decodes the spectrum envelope code S5 to obtain the spectrum envelope, calculates the distance D between the spectrum envelope of the present frame and the spectrum envelope of the immediately preceding frame, and when the distance D is large. Sets a large weighting coefficient α to bring the shaped representative sound source closer to an impulse.

The representative excitation repetition unit 21 repeats the shaped representative excitation S26 obtained by the excitation shaping unit 25 at a pitch cycle interval to obtain a decoded excitation signal S22.

FIG. 3 shows an example of the decoded excitation signal S22 generated in this embodiment. FIG. 5A shows the shaped representative excitation and FIG. 5B shows the decoded excitation signal. There is. As shown in the figure, by the processing of the excitation shaping unit 21, a decoded excitation signal with a smooth change between frames can be obtained.

(Second Embodiment) In the first embodiment described above,
The sound source shaping unit 25 uses the distance value D between the spectral envelopes of the current frame and the immediately preceding frame to set the magnitude of the weighting coefficient α, but the present invention is not limited to this.
It may be configured as in the second embodiment using the cross-correlation values of the representative sound sources of both frames.

The present invention is not limited to the above embodiment. The above-described embodiment is an exemplification, and has substantially the same configuration as the technical idea described in the claims of the present invention, and any device having the same function and effect can be realized by the present invention. It is included in the technical scope of the invention.

[0026]

As described above, according to the present invention, since the representative sound source is shaped by the sound source shaping means so as to be close to an impulse, a decoded sound source signal having a smooth change between frames can be obtained. Further, when the change of the input voice signal is large, the shaping is performed so as to be closer to the impulse, so that the decoded excitation signal that can follow the large change can be obtained. Therefore, there is an advantage that a decoded voice signal of good quality can be obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a speech coding / decoding apparatus that is a first embodiment of the present invention.

[Fig. 2] Fig. 2 is a diagram for explaining the operation of a sound source shaping unit in the speech coding / decoding device shown in Fig. 1.

[Fig. 3] Fig. 3 is a diagram for describing an operation of generating a decoded excitation signal in the speech coding / decoding device shown in Fig. 1.

FIG. 4 is a block diagram showing a configuration of a conventional speech encoding / decoding device.

FIG. 5 is a diagram for explaining an operation of generating a decoded excitation signal in the speech coding / decoding apparatus shown in FIG.

[Explanation of symbols]

 2 spectrum envelope analysis unit 4 inverse filter unit 7 phase equalization unit 9 representative sound source extraction unit 11 pitch period extraction unit 13 voiced / unvoiced judgment unit 17 representative sound source encoding unit 19 representative sound source decoding unit 21 representative sound source repetition unit 23 synthesis filter Part 25 Excitation shaping part 101 Speech coding / decoding device 110 Coding part 120 Decoding part 200 Speech coding / decoding device 210 Coding part 220 Decoding part S Signal etc.

Claims (2)

[Claims] 1. An input speech signal is analyzed for each frame of a predetermined length to obtain spectral envelope information, a sound source signal and a pitch period of the input speech signal, and when the input speech signal is a voiced sound. A representative sound source for one pitch cycle length is extracted after performing a short-time phase equalization process so that the phase of the sound source signal becomes zero at the maximum amplitude position on the signal waveform appearing at intervals of the pitch cycle. An encoding unit that encodes the excitation, the pitch period, and the spectrum envelope information and outputs the spectrum envelope code, and decodes the encoded parameters output from the encoding unit,
When the input speech signal is a voiced sound, a decoded excitation signal is generated by arranging the decoded representative excitations of one pitch period length at intervals of the pitch period,
In a speech coding / decoding apparatus comprising: a decoding unit that decodes a decoded speech signal using this decoded excitation signal and a decoded spectrum envelope, the representative excitation and impulse signal of the corresponding decoded frame And a representative excitation repeating unit that generates the decoded excitation signal by arranging shaped representative excitations output from the excitation shaping unit at intervals of the pitch cycle. A speech encoding / decoding device characterized by the above. 2. The value of the weighting coefficient when the sound source shaping means weights and adds the representative sound source and the impulse signal according to the magnitude of change in the audio signal of the corresponding frame and a frame in the vicinity of the corresponding frame. The speech coding / decoding apparatus according to claim 1, wherein