JP3428594B2 - Audio encoding device, audio decoding device, audio encoding method, and audio decoding method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding / decoding device used in a voice codec for digital mobile communication and a voice synthesizer for voice output of various devices.

[0002]

In the field of digital mobile communications,
In order to cope with the increase in the number of subscribers, there is a demand for a lower bit rate speech coding method, and research and development are being conducted by each research institute. And now the bit rate is 8 kb
ps is standardized, and each research institute has 4.8 kbps
We are conducting research and development towards low bit rates.
However, a speech encoding / decoding device having sufficient performance has not yet been obtained.

The technique for realizing a low bit rate is described in the following 2
It can be divided into street ways.

The first method is, as represented by CELP, a method in which a speech in a certain analysis section is encoded by an LPC system parameter and a driving sound source, and a pitch component is produced by a long-term prediction filter. This method is currently the most efficient method for medium bit rates from 6.7kbps to 16kbps. However, at 4 kbps or less, the sound quality is considerably degraded, and it is difficult to realize a low bit rate.

The second method is a method in which most of human voice is voiced sound having a fundamental frequency, and pitch analysis is performed to transmit only the pitch and one pitch waveform. This method is quite effective in encoding below 6 kbps, and can be said to be the most effective method for realizing a low bit rate. However, an error always occurs in pitch detection.
Also, the amount of calculation required for pitch detection is quite large, and it can be said that pitch detection in real time is difficult.

Therefore, it has been considered difficult to realize a low bit rate for the above reasons.

[0007]

As described above, it has been difficult to realize a low bit rate with the conventional voice encoding / decoding technology. This is because long-term prediction does not provide sufficient sound quality, and pitch synchronization makes it difficult to detect pitch.

In view of the above problems, the present invention realizes a low bit rate by combining the advantages of the two methods and by using long-term prediction for pitch-synchronous coding to prevent pitch errors. With the goal.

[0009]

In order to achieve this object, the present invention provides an A / D converter for converting an input voice into a digital voice signal and a synthesized waveform storage for storing past synthesized waveforms. And a selected combined waveform selected based on a pitch for long-term prediction from a partial section of the digital audio signal of a predetermined time length and the past combined waveform as an input, the selected combined waveform, a long-term prediction coefficient, The selected synthetic waveform and the long-term prediction for minimizing the error power between the pitch and the basic waveform for the long-term prediction and the new synthetic waveform formed by using the length and the partial section of the digital speech signal. A coefficient, a pitch for the long-term prediction, an acoustic analysis unit that determines the basic waveform and its length, and a basic waveform encoding that encodes the basic waveform obtained by the acoustic analysis unit. And creating a synthetic waveform based on the long-term prediction coefficient and the pitch for the long-term prediction obtained by the acoustic analysis unit, and the code of the basic waveform and the code of the length thereof obtained by the basic waveform coding unit. A speech coder having a speech synthesizing section for outputting to the synthesized waveform storing section, and outputting the long-term prediction coefficient, the pitch for the long-term prediction, the code of the basic waveform and the code of the length thereof. .

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The invention according to claim 1 of the present invention is an A / D converter for converting an input voice into a digital voice signal, a synthesized waveform storage for storing past synthesized waveforms, With a selected synthesized waveform selected based on a pitch for long-term prediction from a partial section of the digital speech signal of a predetermined time length and the past synthesized waveform as an input,
In order to minimize the error power between the selected synthesized waveform, the long-term prediction coefficient, the pitch for the long-term prediction, the basic waveform and the new synthesized waveform formed by using the length and the partial section of the digital audio signal. , The selected combined waveform,
The long-term prediction coefficient, a pitch for the long-term prediction, an acoustic analysis unit that determines the basic waveform and its length, a basic waveform encoding unit that encodes the basic waveform obtained by the acoustic analysis unit, the acoustic A synthetic waveform is created based on the long-term prediction coefficient and the pitch for the long-term prediction obtained by the analysis unit, the code of the basic waveform and the code of the length thereof obtained by the basic waveform coding unit, and the synthesized waveform is stored. And a speech synthesizing unit for outputting to the unit, and outputs the long-term prediction coefficient, the pitch for the long-term prediction, the code of the basic waveform, and the code of the length thereof.

With the above configuration, a low bit rate can be realized by encoding a 1-pitch basic waveform, and pitch error can be prevented because long-term prediction is used. Therefore, low bit rate speech coding can be performed efficiently.

According to a second aspect of the present invention, the synthesized waveform is obtained by multiplying the selected synthesized waveform based on the pitch for long-term prediction by the long-term prediction coefficient for each length of the basic waveform, and decoding the result to obtain the decoded basic waveform. The speech coding apparatus according to claim 1, wherein the result of summing the waveforms is added up to the length of the analysis section, and a low bit rate can be realized by coding the basic waveform of one pitch. Pitch errors can be prevented because long term prediction is used. Therefore, low bit rate speech coding can be performed efficiently.

According to a third aspect of the present invention, a basic waveform decoding unit that decodes the basic waveform based on the input code of the basic waveform and the code of the length of the basic waveform, a long-term prediction coefficient, and a long-term prediction And a speech waveform decoding section for decoding a digital speech signal having a predetermined time length by using the pitch for decoding and the decoded basic waveform output from the basic waveform decoding section and its length as input, A voice decoding device having a D / A conversion unit for converting a decoded digital voice signal output from a voice waveform decoding unit into an analog voice signal, and efficiently performing low bit rate voice decoding. You can

According to a fourth aspect of the present invention, a basic waveform decoding unit for decoding the basic waveform based on the input code of the basic waveform and the code of the length of the basic waveform, a long-term prediction coefficient, and a long-term prediction And a speech waveform decoding section for decoding a digital speech signal of a predetermined time length, using the pitch for decoding and the decoded basic waveform output from the basic waveform decoding section and its length as input, A voice decoding device having a D / A conversion unit for converting a decoded digital voice signal output from a voice waveform decoding unit into an analog voice signal, and efficiently performing low bit rate voice encoding. You can

According to a fifth aspect of the present invention, in the synthesized waveform, the selected synthesized waveform based on the pitch for long-term prediction is multiplied by the long-term prediction coefficient for each length of the basic waveform, and the result is the decoded basic waveform. The speech coding method according to claim 4, wherein the result of summing the waveforms is summed up to the length of the analysis section, and speech coding at a low bit rate can be efficiently performed.

According to a sixth aspect of the present invention, there is provided a basic waveform decoding step of decoding the basic waveform based on the input basic waveform code and the basic waveform length code, and a long-term prediction coefficient,
A speech waveform decoding step for decoding a digital speech signal having a predetermined time length by inputting the pitch for long-term prediction, the decoded basic waveform obtained in the basic waveform decoding step and its length. And a D / A conversion step of converting the decoded digital audio signal obtained in the audio waveform decoding step into an analog audio signal, which is a low bit rate audio decoding method. You can do it well.

(Embodiment 1) Hereinafter, a first embodiment of the present invention will be described with reference to the drawings.

In FIG. 1, 1 is an A / D conversion section, 2 is a synthesized waveform storage section, 3 is an acoustic analysis section, 4 is a basic waveform coding section, 5 is a speech synthesis section, 6 is a basic waveform decoding section, Reference numeral 7 is a speech waveform decoding unit, 8 is a D / A conversion unit, 10 is an encoder, 11
Is a decoder.

Next, the operation of the speech encoding / decoding apparatus according to the first embodiment of the present invention will be described.

First, the function of the encoder 10 will be described with reference to FIG. The A / D converter 1 converts an input voice input from a microphone (not shown) into a digital signal. Next, in the acoustic analysis unit 3, a voice signal of a fixed time is taken into a RAM memory (not shown), and the correlation between this analysis section and the synthetic waveform stored in the synthetic waveform storage unit 2 is analyzed.
The pitch analysis of the analysis section is performed at the same time, and the position of the partial section of the correlated waveform and the basic waveform of one pitch are obtained. For the extraction method of this parameter and basic waveform,
The details will be described later.

Further, in the basic waveform encoding unit 4,
The basic waveform obtained in the acoustic analysis unit 3 is encoded.
As a specific method, a method of matching the phases and performing VQ as a waveform, a method of converting into a frequency domain and then encoding, and the like can be mentioned.

Here, in the voice synthesizing section 5, the basic waveform of one pitch is decoded based on the code obtained in the basic waveform coding section 4, and the long-term prediction coefficient extracted in the acoustic analysis section 3 and the above basic waveform. Decrypt using and
The obtained composite waveform is stored in the composite waveform storage unit 2. The function of the speech synthesis unit 5 is a combination of the functions of the basic waveform decoding unit 6 and the speech waveform decoding unit 7 in the decoder 11, and the details thereof will be described when the decoder 11 is described.

Here, the method of extracting the parameters and the basic waveform in the acoustic analysis unit 3 will be described in detail. The synthesis formula of the present invention is shown in (Equation 1).

[0024]

[Equation 1]

In this (Equation 1), n is the pitch section number,
q is a pitch period, β is a long-term prediction coefficient, p is a long-term prediction pitch, Xnq + i and Xnq + ip are all synthetic waveforms, and Yi is a decoded basic pitch of one pitch.

Therefore, p, q, β and Yi are calculated so that the waveform synthesized by this synthesis formula is closest to the original waveform.

Now, given p and q, the original waveform

[0028]

[Equation 2]

The error power between the composite waveform of and (Formula 1) is as shown in (Formula 3) below.

[0030]

[Equation 3]

Here, E is the error power, and M is the number of pitch periods in one analysis section. Therefore, it is used that when E is the minimum, the values differentiated by β and Yi are both 0. First, by differentiating with β, (Equation 4) is obtained.

[0032]

[Equation 4]

Therefore, by using the following (Equation 5) and (Equation 6) to simplify the equation and solve for β, (Equation 7) is obtained.

[0034]

[Equation 5]

[0035]

[Equation 6]

[0036]

[Equation 7]

On the other hand, when (Equation 3) is differentiated by Yk, (Equation 8) is obtained.

[0038]

[Equation 8]

Then, by substituting (Equation 8) into (Equation 7), β
Is obtained, and each value is used to obtain each Yk. This is performed for all p and q, the error E is evaluated, and p and q with the smallest error are selected.

However, a full search for these p and q requires a large amount of calculation. As a method of reducing the calculation amount, there are a method of narrowing down the candidates of p and q by the value of Vp and searching, a method of setting q to the value of p at the maximum Vp, and the like.

Next, the function of the decoder 11 will be described with reference to FIG. First, in the basic waveform decoding unit 6, 1
Synthesize the pitch basic waveform. Then, the speech waveform decoding unit 7 uses the 1-pitch basic waveform synthesized by the basic waveform decoding unit 6 and the long-term prediction coefficient to generate a speech waveform of one analysis section based on the synthesis formula (Equation 1). To synthesize.
Then, the D / A converter 8 converts the analog signal and outputs the analog signal.

In order to verify the coding effect of the speech coding / decoding apparatus of the present invention, a preliminary simulation experiment of speech coding / decoding was conducted. The voice for evaluation is "The explosion sound spreads to the plateau of the silver world" uttered by one man, and it was coded with a sampling rate of 8 kHz and 12-bit PCM. In the simulation, the basic waveform coding
It was used as it was without decoding or scalar encoding of β. Both p and q are coded with 7 bits. As a result, a segmental S / N ratio of 13.75 dB was obtained,
There was no pitch error. In addition, 1 pitch waveform is 14
Considering that the coding can be performed with ˜20 dB, the coding can be performed with the S / N ratio of 9 to 12 dB. Therefore, it is possible to achieve the above object.

[0043]

As described above, according to the present invention, since the pitch error is prevented from occurring by using the long-term prediction for the pitch synchronization encoding, the voice is encoded at the low bit rate without causing the pitch error. And can be decrypted.

[Brief description of drawings]

FIG. 1 is a speech coding / encoding method according to a first embodiment of the present invention.
Block diagram of decryption device

[Explanation of symbols]

1 A / D converter 2 Synthetic waveform storage 3 Acoustic analysis section 4 Basic waveform encoder 5 Speech synthesizer 6 Basic waveform decoding section 7 Speech waveform decoding section 8 D / A converter 9 transmission lines 10 encoder 11 Decoder

─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-5-73098 (JP, A) JP-A 8-69300 (JP, A) JP-A 1-205199 (JP, A) JP-A 62- 135899 (JP, A) JP-A 1-126700 (JP, A) JP-A 2-8900 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G10L 11/04 G10L 13 / 00 G10L 19/00 G10L 19/04

Claims (6)

(57) [Claims] 1. An A / D converter for converting an input speech into a digital speech signal, a synthesized waveform storage section for storing past synthesized waveforms, and a partial section of the digital speech signal of a predetermined time length. And using the selected combined waveform selected from the past combined waveform based on the pitch for long-term prediction as input, the selected combined waveform, the long-term prediction coefficient, the pitch for the long-term prediction, the basic waveform and its length are used. The selected synthetic waveform, the long-term prediction coefficient, the pitch for the long-term prediction, and the basic waveform and its length for minimizing the error power between the new synthetic waveform formed by the above-mentioned method and the partial section of the digital speech signal. An acoustic analysis unit that determines the depth, a basic waveform encoding unit that encodes the basic waveform obtained by the acoustic analysis unit, the long-term prediction coefficient and the previous obtained by the acoustic analysis unit The speech synthesis unit that creates a synthetic waveform based on the pitch for long-term prediction, the code of the basic waveform obtained by the basic waveform encoding unit, and the code of the length thereof and outputs the synthesized waveform to the synthetic waveform storage unit. A speech coding apparatus that outputs the long-term prediction coefficient, the pitch for the long-term prediction, the code of the basic waveform, and the code of the length thereof. 2. A synthesized waveform is obtained by multiplying a pitch-based selected synthesized waveform for long-term prediction by a long-term prediction coefficient for each length of the basic waveform, and adding the decoded basic waveform to the result. The speech coding apparatus according to claim 1, wherein the speech coding apparatus is obtained by adding up to a length of an analysis section. 3. A basic waveform decoding unit for decoding a basic waveform based on the input basic waveform code and basic waveform length code, a long-term prediction coefficient, a pitch for long-term prediction, and the basic waveform. An output of the speech waveform decoding unit, and a speech waveform decoding unit that receives the decoded basic waveform that is the output of the decoding unit and the length thereof and decodes a digital speech signal of a predetermined time length. D / A for converting a decoded digital voice signal that is an analog voice signal
A speech decoding apparatus having a conversion unit. 4. An A / D conversion process for converting an input voice into a digital voice signal, and a partial interval of the digital voice signal having a predetermined time length and a pitch for long-term prediction from a past synthesized waveform. A selected synthesized waveform selected as an input, a new synthesized waveform formed using the selected synthesized waveform, the long-term prediction coefficient, the pitch for the long-term prediction, the basic waveform and the length thereof, and a partial section of the digital voice signal. Of the selected combined waveform, the long-term prediction coefficient, the pitch for the long-term prediction, and the basic waveform and its length for minimizing the error power of A basic waveform encoding step of encoding a basic waveform, the long-term prediction coefficient obtained in the acoustic analysis step, a pitch for the long-term prediction, and the basic waveform It has a voice synthesizing step of creating a synthesized waveform based on the code of the basic waveform and the code of its length obtained by the encoding section, and a synthetic waveform storing step of storing the synthesized waveform as a past synthesized waveform. A speech coding method which outputs a prediction coefficient, a pitch for the long-term prediction, a code of the basic waveform, and a code of the length thereof. 5. The synthesized waveform is obtained by multiplying a pitch-based selected synthesized waveform for long-term prediction by a long-term prediction coefficient for each length of the basic waveform, and adding the decoded basic waveform to the result. The speech coding method according to claim 4, wherein the speech coding method is obtained by adding up to a length of an analysis section. 6. A basic waveform decoding step of decoding a basic waveform based on a code of the input basic waveform and a code of the length of the basic waveform, a long-term prediction coefficient, a pitch for long-term prediction, and the basic waveform. The decoded basic waveform obtained in the decoding step and its length are input, and a speech waveform decoding step of decoding a digital speech signal of a predetermined time length and a speech waveform decoding step D / that converts the decoded digital audio signal to an analog audio signal
A voice decoding method having an A conversion step.