JPH0675599A - Voice coding method, voice decoding method and voice coding and decoding device - Google Patents

Abstract

PURPOSE:To improve the degree of clearness of reproduced voice by making the sound source power value of the voice signal during a no voice condition to be no sound subframe sound source power value obtained for each serialized plural subframe. CONSTITUTION:Sampled input voice a1 is divided by a framing device 11 and becomes b1 and is processed for every frame. If there is a no voice frame, a switch 35 is controlled and prediction difference signal d1 is sent to a subframing device 37 and for example, is divided to four subframes. A subframe power computer 38 computes the power of signal K3, which is divided into subframes, for every subframe. A subframe power synthesizer 39 places the power obtained for every subframe in a serial manner, makes a sound source power information m3, which is the parameter equivalent to one frame, and outputs it through a switch 40. Thus, an amplitude gain is adjusted for every subframe and the amplitude change of no sound region in the reproduced voice is more closely approximated to the input voice.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding method and a voice decoding method suitable for unvoiced sound synthesis, and an apparatus thereof.

[0002]

2. Description of the Related Art Low bit rate (about 2.4 kbps)
Voice coding and decoding device (that is, voice coding and decoding device)
In general, the analysis and synthesis method is used. An LPC vocoder (speech code decoding method based on linear prediction analysis) is the most typical analysis and synthesis method. In this method, the prediction residual signal is pulse train or noise modeled and information is compressed. When the input signal is in the voiced section, pulse train is used, and when the input speech is in the unvoiced section, white noise is used. A speech coding apparatus and a speech decoding apparatus in this conventional example are shown in FIG. 2 and FIG. 3, respectively. In the following description, it is assumed that the speech coding rate is 2.4 kbps and that LSP (line spectrum pair) is used as the linear prediction coefficient (spectral envelope information).

In the speech coding apparatus, for example, the input speech a1 sampled at 8 kHz is processed by the framer 11
One frame is divided into 20 msec and becomes b1, and thereafter each frame is processed. b1 is subjected to linear prediction analysis by the linear prediction analyzer 12, and as a result, the linear prediction coefficient c
1, which is converted into an LSP coefficient, vector-quantized with 20 bits, and transferred to the decoding apparatus of FIG. The linear prediction analysis filter 13 filters b1 using the linear prediction coefficient c1 as a coefficient and outputs the prediction residual signal d1. The voiced / unvoiced decision unit 14 decides whether the input voice is voice or unvoiced, and sends the result e1 to the pitch period extractor 15. The pitch period extractor 15 calculates and outputs the pitch period f1 from the prediction residual signal d1, but when the voiced / unvoiced determination result e1 is unvoiced, the pitch period f1 is calculated.
Voiced / unvoiced information is obtained by inserting and outputting zero. The sound source power calculator 16 calculates and outputs power information (power value) for each frame of the prediction residual signal d1.

Here, the voiced / unvoiced decision method of the LPC vocoder in the decision unit 14 is as follows. In the LPC vocoder, the voiced / unvoiced determination is determined by the degree of periodicity of the sound source (prediction residual signal). The degree of periodicity is determined by the prediction residual signal correlation t _p at a time delay (t = t _p ) corresponding to the pitch period t _p of the input speech signal and the prediction residual signal correlation R at no time delay (t = 0). Check with the value normalized by ₀ . That is, the voiced / unvoiced determination method is performed by the following conditional expression. Voiced section if R _tp / R ₀ ≧ 0.25 Voiceless section if R _tp / R ₀ <0.25

In the speech decoding apparatus of FIG. 3, in order to generate an excitation signal, a pulse train b2 having a pitch period a2 transmitted by the pulse generator 21 is generated,
The white noise generator 22 generates white noise c2.
The pulse gain calculator 25 uses the transmitted sound source power information d2 to obtain an appropriate pulse gain f2 (where the sound source power value is γ and the pitch period is T, the pulse amplitude is (Tγ) ^1/2
Adjust so that the noise gain calculator 2
In step 6, an appropriate noise gain g2 (adjusted so that the transmitted sound source power d2 and the white noise power i2 are the same) is calculated from the transmitted sound source power information d2. If the pitch cycle a2 is zero, the control circuit 23 determines that it is a silent frame, and switches the switches 24 and 27 to the white noise generator 22 side and the noise gain calculator 26 side, respectively, and determines that the pitch cycle a2 is other than zero. For example, assuming that it is a voiced frame, the switches 24 and 27 are switched to the pulse generator 21 side and the pulse gain calculator 25 side, respectively. The multiplier 29 multiplies the selected sound source signal i2 and the selected gain h2, and outputs the gain adjusted sound source signal j2. The linear synthesis filter 28 filters the linear prediction coefficient k2 as a filter coefficient to generate a reproduced voice j3.

[0006]

FIG. 4 shows the bit allocation of one frame (20 ms, 48 bits) in the conventional method. 1 bit for synchronization and 7 bits for pitch period
Bits, 6 bits for sound source power, 20 bits for linear prediction coefficient (LSP coefficient), 14 bits for unused bits (unused bits are used as some auxiliary information bits in a voiced frame) ).

The problem with the conventional method is that one frame is 20 m.
For s, since only one sound source power information is transmitted, it is not possible to reliably reproduce a rapid amplitude change (which occurs in a time width smaller than 20 ms) that occurs in the consonant part of the voice. An example of this is shown in FIGS. (A) is the input voice (original voice), which is the waveform of the generated sound "tsu". Here, the rapid amplitude change in the part surrounded by the dotted circle is important information for expressing the consonant "ts". However, the playback sound of the conventional method is shown (b)
Then, it cannot be expressed and is averaged by the frames to be flat, so that the reproduced sound is perceived as "hu".

An object of the present invention is to improve the clear voice of the reproduced voice by making the amplitude change of the silent section in the reproduced voice closer to the input voice more faithfully when the input voice is an unvoiced sound (consonant part). An object of the present invention is to provide a speech encoding method, a decoding method, and a synthesizer.

[0009]

According to the present invention, a parameter (a pitch period, a sound source power value, a linear prediction coefficient) that can be identified as voiced or unvoiced is extracted from a voice signal, and the extracted parameter is used as a voice signal. In the voice coding method that outputs instead of the signal, the sound source power value when the voice signal is unvoiced is obtained for each of the serialized subframes, and the subframe sound source power value when there is no voice and (Claim 1).

Furthermore, the present invention receives a parameter output by the speech coding method of claim 1, restores a speech signal using the parameter in the case of voiced sound, and unvoiced sound in the case of unvoiced sound. The audio signal is restored using the above parameters including the subframe sound source power value at that time (claim 2).

Further, the present invention comprises a speech coding section and a speech decoding section, and the speech coding section calculates a linear prediction coefficient from a framing device which divides a speech signal into frame units. , A linear prediction analyzer that outputs this,
A linear prediction component filter that obtains a prediction residual signal from the linear prediction coefficient and the framer output, a determiner that determines whether voiced sound or unvoiced sound is obtained from this prediction residual signal, and a determination unit that determines from the prediction residual signal. A pitch period extractor that obtains the pitch period when voiced according to the determination output and outputs the pitch period when unvoiced is zero, and the determination unit calculates the sound source power value by capturing the prediction residual signal at the time of voiced sound determination, A sound source power calculator that outputs this as a sound source power value at the time of voiced sound determination,
A subframe generator that determines the undecided sound when the unvoiced sound determines the prediction residual signal and divides it into subframes, and a subframe power calculator that calculates the excitation power value of the prediction residual signal for each subframe. The subframe power synthesizer for synthesizing the subframe power and the output means for outputting the subframe power synthesizer as a sound source power value at the time of unvoiced sound determination are provided. A control circuit for discriminating between voiced sound and unvoiced sound from the contents of the above, a pulse generator for generating a pulse synchronized with the pitch period of the voiced sound from the pitch period of the pitch period extractor, a white noise generator, and the above control When the circuit judges voiced sound, it takes in the sound source power value of the voice coding unit in voiced sound, calculates the pulse gain, and multiplies the calculated value by the pulse generator output. When the unvoiced sound is determined by the control circuit, the sound source power value of the unvoiced sound of the speech encoding unit is taken in to calculate the subframe noise gain, and the calculated value is multiplied by the output of the white noise generator. The second means for performing and the linear predictive synthesis filter which takes in the output of the first means at the time of voiced sound and performs the linear predictive synthesis at the time of unvoiced sound determination, and performs the linear predictive synthesis to obtain the reproduced voice. And (claim 3).

[0012]

According to the present invention, a voice signal is coded with a parameter consisting of a pitch period, a sound source power value, and a linear prediction coefficient, and at the time of unvoiced sound, the sound source power value obtained for each subframe is output so that the unvoiced sound is generated. To achieve the encoding (claim 1).

Further, according to the present invention, when unvoiced when receiving and restoring the above parameters, the amplitude of white noise, which is a sound source, is adjusted for each subframe to more faithfully change the amplitude of the unvoiced signal. (Claim 2).

Further, the present invention is provided with an encoding unit for transmitting the unvoiced sound source power value of each subframe as a parameter, and a decoding unit for receiving and restoring the same so that the amplitude change in the unvoiced section is more faithful. It enables synthesis close to the input signal (Claim 3).

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows the configuration of a speech coding apparatus of the present invention, and FIG. 6 shows the configuration of a speech decoding apparatus. Here, it is assumed that the speech coding rate is 2.4 kbps and that the LSP (line spectrum pair) is used as the linear prediction coefficient unit (spectral envelope information).

In FIG. 1, the parts newly added in comparison with FIG. 2 are a switch 35, a subframe converter 37, a subframe power calculator 38, a subframe power combiner 39,
The switch 40. In FIG. 6, the parts newly added in comparison with FIG. 3 are switches 44 and 47, and in addition, the calculator 46 calculates the subframe noise gain.
Different from

In the speech coder, for example, the input speech a1 sampled at 8 kHz is processed by the framer 11
One frame is divided into 20 msec and becomes b1, and thereafter each frame is processed. b1 is subjected to linear prediction analysis by the linear prediction analyzer 12, and as a result, the linear prediction coefficient c
1, which is converted into LSP coefficients, vector-quantized with 20 bits, and transmitted to the decoder. The linear prediction analysis filter 13 sets the linear prediction coefficient c1 as a coefficient to b1.
To output a prediction residual signal d1. The voiced / unvoiced decision unit 14 decides whether the input voice is voiced or unvoiced, and the result e1 is determined as the pitch period extractor 15
Send to. In the pitch period extractor 15, the prediction residual signal d1
Then, the pitch period f3 is calculated and output, but when the voiced / unvoiced determination result e1 is unvoiced, zero is inserted in the pitch period f3 and output to obtain voiced / unvoiced information. In the case of a voiced frame, the sound source power calculator 16 controls the switch 35 to calculate power information of each frame of the prediction residual signal d1 and outputs the power information via the switch 40.

In the case of an unvoiced frame, the switch 35 is controlled to send the prediction residual signal d1 to the subframe converter 37, which divides it into, for example, four subframes (5 ms). The subframe power calculator 38 calculates the power 14 for each subframe for the signal k3 divided into subframes. The subframe power combiner 39 serially arranges the power obtained for each subframe to create sound source power information m3 that is a parameter for one frame, and outputs this through the switch 40. The switches 35 and 40 correspond to the sound source power calculator 36 side and the subframe converter 37 / subframe power calculator 3 according to the voiced / unvoiced information e1.
8. Switch the subframe power combiner 39 side.

In the speech decoding apparatus shown in FIG. 6, in order to generate a sound source signal, the pulse generator 21 generates a pulse train b4 having a period of the transmitted pitch period a4, and the white noise generator 22 causes the white noise generator 22 to generate white noise. c4 is generated. In the pulse gain calculator 25, the transmitted sound source power information f4 is transmitted.
Then, an appropriate pulse gain g4 (adjusted so that the pulse amplitude is (Tγ) ^1/2 where γ is the sound source power and T is the pitch period) is calculated by the subframe noise gain calculator 46. From the transmitted sound source power information f4, an appropriate noise gain h4 of each subframe (adjusted so that the sound source power f4 and the white noise power e4 are the same) is calculated, and each gain is output for each subframe. To do. If the pitch cycle a4 is zero, the control circuit 23 determines that the frame is a silent frame, and the switch 44 and the switch 47.
Are switched to the white noise generator 22 side and the sub-frame noise gain calculator 46 side respectively, and if the pitch period a4 is not zero, it is determined that it is a voiced frame, and the switches 44 and 47 are respectively set to the pulse generator 21 side and the pulse gain. Switch to the calculator 25 side. The multiplier 29 multiplies the selected sound source signal e4 by the selected gain i4 and outputs the gain adjusted sound source signal j4. The linear synthesis filter 28 uses the linear prediction coefficient k4 as a filter coefficient j
4 is filtered and reproduced voice j5 is generated.

The bit allocation of one frame (20 ms, 48 bits) according to the present invention is shown in FIG. FIG. 7A shows the conventional example of FIG. 4 for comparison. 1 bit for synchronization, 7 bits for pitch period, 16 bits for sound source power (4 bits / subframe: 1 subframe, the power fluctuation range is smaller than that of the whole frame. Therefore, it is possible to quantize with 4 bits), and 2 bits for the linear prediction coefficient (LSP coefficient).
There are 0 bits and 4 unused bits.

In the present invention, since the sound source power information is transmitted every 5 ms of one sub-frame, compared with the prior art, the abrupt amplitude change (occurs in a time width smaller than 20 ms) occurring in the consonant part of the voice. ) Can be handled more reliably.
This example is shown in FIGS. 8A, 8B, and 8C. (A) is the input voice (original voice), which is the waveform of the generated sound "tsu". Here, the rapid amplitude change in the part surrounded by the dotted circle is important information for expressing the consonant "ts". That is, the frame is divided into subframes 1, 2,
Considering sections 3 and 4, subframe 2 has a pulsed waveform, and therefore has higher power than other subframes. This pulse-shaped waveform is an important element for expressing "ts". As described above, in the case of (b) which shows the reproduced sound of the conventional method, it cannot be expressed and becomes flat, so that the reproduced sound is perceived as "hu" in the sense of hearing.
(C) is a reproduced speech waveform according to the present invention, and a sudden change in amplitude can be expressed, and can be perceived as "tsu". That is, since the electric power is sent for each sub-frame, the electric power in the sub-frame 2 can be expressed more than in the other sub-frames, and "ts" can be heard in the sense of hearing.

When used in this embodiment, the single syllable intelligibility (one of the objective evaluation methods for reproduced voices, that is, the correct answer rate [%] obtained by arranging various syllables at random and performing a listening test, (Most of the hearing loss occurs in the consonant part), which is 50.6% in the conventional method, while 55.
It was improved to 6% and it was confirmed that the quality of the reproduced voice was improved.

According to the present invention, a frame is decomposed into subframes for a sound source signal, and an amplitude gain adjustment is performed for each subframe, so that a change in amplitude of a voiceless section in a reproduced voice can be faithfully reproduced. It has become possible to improve the clear voice of the reproduced voice.

[Brief description of drawings]

FIG. 1 is a diagram showing an embodiment of a speech coder according to the present invention.

FIG. 2 is a diagram showing a conventional speech encoding device.

FIG. 3 is a diagram showing a conventional speech decoding device.

FIG. 4 is a diagram showing a conventional parameter transmission format.

FIG. 5 is a diagram showing a state of conventional audio reproduction.

FIG. 6 is a diagram showing an embodiment of a speech decoding apparatus of the present invention.

FIG. 7 is a diagram showing a parameter transmission format of the present invention in comparison with a parameter transmission format of a conventional example.

FIG. 8 is a diagram showing a state of audio reproduction of the present invention in comparison with a conventional example of audio reproduction.

[Explanation of symbols]

 11 Framer 12 Linear Prediction Analyzer 13 Linear Prediction Analysis Filter 14 Voiced / Unvoiced Determinator 15 Pitch Period Extractor 16 Source Power Calculator 21 Pulse Generator 22 White Noise Generator 23 Control Circuit 25 Pulse Gain Calculator 28 Linear Prediction Synthesis filter 35, 40 Switch 37 Subframe converter 38 Subframe power calculator 39 Subframe power combiner 44, 47 Switch 46 Subframe noise gain calculator

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Masayasu Miyake 2-3-3 Toranomon, Minato-ku, Tokyo Kokusai Electric Co., Ltd.

Claims (3)

[Claims] 1. A voice code for extracting a parameter (a pitch period, a sound source power value, a linear prediction coefficient, which can be identified as voiced or unvoiced) from a voice signal and outputting the extracted parameter instead of the voice signal. In the speech coding method, the sound source power value when the sound signal is unvoiced is a subframe sound source power value in the absence of sound, which is obtained for each of a plurality of serialized subframes. 2. A parameter output by the voice encoding method according to claim 1 is received, and in the case of voiced sound, the voice signal is restored using the parameter, and in the case of unvoiced sound, the sub-parameter at that time is restored. A speech decoding method that restores a speech signal by using the above parameters including the frame sound source power value. 3. A speech coding unit and a speech decoding unit, wherein the speech coding unit divides a speech signal into frame units, and a linear prediction coefficient is calculated from this output, and the linear prediction coefficient is calculated. A linear prediction analyzer that outputs, a linear prediction component filter that obtains a prediction residual signal from this linear prediction coefficient and the framer output, a determiner that determines whether this is a voiced sound or an unvoiced sound from the prediction residual signal, and A pitch period extractor for obtaining the pitch period in the voiced state and outputting the pitch period in the unvoiced state to zero according to the determination output of the determination unit from the prediction residual signal,
The sound source power calculator that the decision unit takes in the above prediction residual signal at the time of voiced sound judgment and calculates the sound source power value, and outputs this as the sound source power value at the time of voiced sound judgment, and the above prediction residual error at the time of the unvoiced sound judgment by the decision unit. A subframe converter that takes in a signal and divides it into subframes, a subframe power calculator that calculates the excitation power value of the prediction residual signal for each subframe, and a subframe power combiner that combines this subframe power And a means for outputting this subframe power combiner as a sound source power value when determining unvoiced sound, and the speech decoding unit distinguishes between voiced sound and unvoiced sound from the content of the pitch cycle of the pitch cycle extractor. A control circuit, a pulse generator that generates a pulse synchronized with the pitch period of voiced sound from the pitch period of the pitch period extractor, a white noise generator, and When the control circuit determines the voiced sound, the first means for taking in the sound source power value of the voice encoding unit during the voiced sound, calculating the pulse gain, and multiplying the calculated value by the pulse generator output; When the unvoiced sound is determined, the sound source power value of the unvoiced sound of the speech encoding unit is taken in to calculate the subframe noise gain, and the second means for multiplying the calculated value and the output of the white noise generator, A linear predictive synthesis filter that takes in the output of the first means, performs linear predictive synthesis, and takes the output of the second means when performing unvoiced sound determination, and performs linear predictive synthesis, thus obtaining reproduced speech;
A voice codec, which comprises: