JP2832942B2 - Multi-pulse encoder - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a multi-pulse coding apparatus, and more particularly, to a pitch prediction type multi-pulse coding apparatus which can easily search for pulses and has a simple apparatus.

[Conventional technology]

It is known that a speech waveform has two linear predictabilities, that is, a so-called proximity predictability and a pitch predictability. Early multi-pulse encoding devices were configured to encode speech using only the proximity predictability among them.

In a conventional non-pitch prediction type multi-pulse coding apparatus using only proximity predictability, a multi-pulse is determined via a cross-correlation coefficient between an impulse response of a proximity prediction filter and a speech waveform to be analyzed. In recent years, in order to increase the coding efficiency of a multi-pulse coding device, attempts have been made to use pitch prediction in this device.

Conventionally, in a multi-pulse coding apparatus that performs this kind of pitch prediction, speech is synthesized using a cascade connection of a proximity prediction filter and a pitch prediction filter by linear prediction analysis (hereinafter, referred to as LPC analysis). . Therefore, this 2
The impulse response of two filters connected in cascade is calculated, the cross-correlation coefficient is calculated in place of the impulse response of the proximity prediction filter in the conventional device, and multi-pulse search is performed.

Further, in the conventional pitch prediction type multi-pulse encoding method, the substantial duration of the impulse response is long, for example,
Since it can reach several times to ten times the frame length of 20 ms, there is a point that it does not match with the multi-pulse coding.

[Problems to be solved by the invention]

The conventional multi-pulse coding apparatus described above has a drawback that the number of operations is significantly increased because a proximity filter and a pitch prediction filter based on LPC analysis are used in cascade in searching for a multi-pulse.

[Means for solving the problem]

A multi-pulse coding apparatus according to the present invention is a multi-pulse coding apparatus that transmits, decodes, and outputs an encoded signal obtained by encoding an input audio signal. Pitch extracting means for outputting a signal and a pitch gain coefficient signal; and one of a time domain separated by a pitch cycle from one end of the frame time length by dividing the audio signal by a pitch cycle for each frame time length. An audio signal to be encoded by sequentially performing operations of adding one audio signal in one time domain and an audio signal obtained by multiplying an adjacent audio signal in the previous time domain by a pitch gain coefficient until reaching the other end of the frame time length. And amplitude correction means for correcting the amplitude of the multi-pulse obtained from the audio signal after addition using the audio signal before addition.

〔Example〕

Next, the present invention will be described with reference to the drawings. Although the present embodiment is particularly directed to a multi-pulse encoding device on the premise of spectrum enhancement, this is not related to the essence of the present invention.

FIG. 1 is a block diagram showing the configuration of one embodiment of the present invention,
FIG. 2 is a block diagram showing an example of a spectrum emphasis filter constituting the present invention, and FIGS. 3 (a) and 3 (b) are waveform diagrams of a spectrum-emphasized audio signal showing the operation of pitch addition of the present invention. , (A) is an input voice waveform diagram, (b)
FIG. 6 is an input voice waveform diagram obtained by preprocessing pitch addition of the input voice shown in FIG.

First, the outline of the present invention will be described with reference to FIGS.
This will be described with reference to FIG.

The present invention provides a multi-pulse signal by providing a pitch addition unit that performs linear addition of an audio waveform based on the pitch period of the audio waveform, thereby reducing the number of calculations that can reduce the transmission bandwidth and performing linear prediction, and By quantizing the multi-pulse signal, the configuration is simplified as compared with the related art.

Therefore, focusing on the fact that the approximate period of the input audio signal is substantially constant, the present invention utilizes the pitch period.

However, since two linear predictions, pitch prediction and proximity prediction, are performed, the following method of obtaining a multipulse signal by reducing the number of calculations is employed.

Therefore, in this case, in order to obtain a multi-pulse signal by reducing the number of operations, the audio signal is divided by the pitch period for each frame period, and the audio signals in the time domain separated by the pitch period from the end of the frame are sequentially adjacent. A speech signal obtained by multiplying the speech signal in the time domain on the frame end side by the pitch gain coefficient g is also added, and a linearly added speech signal is output.

A multi-pulse signal is generated from the input audio signal. In other words, as the audio signal is shown in FIG. 3 (a), the audio signal e ₁ of time-domain separated by a pitch period P derived from the pitch period signal in each frame period T ~
e _n are supplied.

Here, as shown in FIG. 3 (b), the linearly added audio signal E1 in the _first time domain at the end of the frame is the same as the audio signal e1 in the _first time domain. The second linearly added audio signal E ₂ from the end of the frame is an audio signal in the same time domain
the same as the sum of the ge ₁ multiplied by the audio signal e ₁ and the pitch gain control factor g of the first time domain from e ₂ and the frame end.
Therefore, this linear sum audio signals E ₁ to E _n is obtained as shown in FIG. 3 are performed in sequence (b).

Next, a spectrum enhancement filter that enhances the spectrum will be described.

In the spectrum emphasis filter, LPC coefficients αli (i = 1, 2,.
... P) is required. Next, the transfer function is obtained from the preset attenuation coefficient γ (determined empirically between 0 and 1) and the LPC coefficient α _i. The filter represented by ## EQU1 ## emphasizes a spectrum having a large amplitude in the spectrum of the input speech signal, and obtains a speech signal close to a line spectrum by spectrum emphasis.

Specifically, as shown in FIG. 2, an input audio signal is supplied to filter elements C _{1 to} C _p of a transfer function Z ⁻¹ connected in cascade through an adder A ₁ , and the output of each element is output. Is applied to multipliers M _{1 to} M _p , where the LPC coefficient α _i is multiplied by a preset attenuation coefficient γ ^{i, respectively,} and the output is added by adder A ₂ , and further added by adder A ₁ It is added to the input audio signal. In such Such connections, the output of the adder A ₁ i.e., the input of the filter elements C ₁ of the transfer function Z ^-1, the output of the spectrum highlighted audio signal.

Next, the configuration and operation of an embodiment of the present invention will be described with reference to FIG.

1, this embodiment includes an LPC analysis unit 1, a spectrum emphasis unit 2, a waveform encoding unit 3, a waveform decoding unit 4, an LPC synthesis unit 5, and an amplitude correction unit 6. Is done. An example of the internal configuration of each component will be described below, and details of each component will be described.

The sampled input audio signal 100 is supplied to a waveform extracting section 11 of the LPC analyzing section 1 and a waveform extracting section 17 of the spectrum emphasizing section 2.
, And the pitch extraction unit 31 of the waveform encoding unit 3.

The LPC analysis means 1 includes a waveform cutout unit 11 and an LPC analysis unit 12. In the waveform cutout unit 11, a signal of 30 ms is cut out from the input audio signal 100 every 20 ms, for example, and the cut out becomes smooth. The frame signal that has passed through the hamming window is output to the LPC analysis unit 12. In the LPC analysis unit 12, the input frame signal is analyzed, and as a result, the LPC coefficient signal 101 up to the maximum order P is input to the K quantization encoding unit 13 of the spectrum emphasis unit 2.

The spectrum enhancement means 2 includes a K quantization / decoding section 13 and a K
-It has an α conversion unit 14, an attenuation coefficient application unit 15, a spectrum emphasis filter 16, and a waveform cutout unit 17.

In the K quantization decoding unit 13 to which the LPC coefficient signal 101 has been input, the quantized LPC coefficient signal 104 is output to the multiplexing / synthesizing unit 35 of the waveform encoding unit 3 and is quantized and decoded. The LPC coefficient signal is output to the K / α conversion unit 14. K
In the α conversion unit 14, the decoded LPC coefficient signal is input, and the LPC coefficient signal representing the LPC coefficient α _i is output to the attenuation coefficient application unit 15.

In the attenuation coefficient applying unit 15 to which the LPC coefficient signal representing the LPC coefficient α _i is input, the product of the attenuation coefficient γ ⁱ and the LPC coefficient α _i empirically selected in advance is calculated and represents α _i · γ ⁱ The signal, that is, the α · γ signal 103 is output to the spectrum emphasizing filter 16 and simultaneously to the impulse response section 20 and the spectrum emphasizing filter 21 of the waveform encoding means 3.

On the other hand, in the waveform extracting unit 17 to which the input audio signal 100 is input, the extracting for each frame is performed in synchronization with the frame signal of the waveform extracting unit 11, for example, every 40 ms, for example, every 40 ms. The cut-out input audio signal is subjected to spectrum emphasis by a spectrum emphasis filter 16 to emphasize the spectrum of a large amplitude in the input audio signal for each frame. It is output to the unit 27 and the maximum value search unit 61 of the amplitude correction means 6.

Next, the waveform encoding means 3 will be described. The waveform encoding unit 3 includes an impulse response unit 20, a spectrum emphasis filter 21, a temporary memory 24, a maximum value search correction unit 25,
Pulse quantizer 26, pitch adder 27, pitch extractor
31, a pitch gain quantization decoding unit 32, and a multiplexing / synthesizing unit 35
And

Α ・ γ expressed by the product of LPC coefficient α _i and damping coefficient γ ⁱ
Signal 103 is output to spectrum emphasis filter 21 and impulse response section 20. In the impulse response section 20, α ·
The impulse signal is input to the LPC synthesis filter determined by the γ signal 103, and the response signal is
Entered into 21. The spectrum emphasis filter 21 is formed with a filter having a transmission characteristic determined by the input α / γ signal 103, and the impulse response signal of the LPC synthesis filter separately input from the impulse response unit 20 is used by this filter. The spectrum is emphasized, and the resulting impulse response signal 110 is output to the maximum value search correction unit 25.

On the other hand, in the pitch extraction unit 31 to which the sampled input audio signal 100 has been input, the pitch of the input audio signal is extracted by, for example, autocorrelation processing, and the pitch period signal 112 indicating this is multiplexed with the pitch addition unit 27. The pitch gain signal representing the pitch gain coefficient (pitch prediction coefficient) is output to the pitch gain quantization / decoding section 32 while being output to the conversion / synthesis section 35. In the pitch gain quantization / decoding section 32, the input pitch gain signal is output as a quantized pitch gain signal 113 to the multiplexing / synthesizing section 35, and the pitch gain signal is quantized and decoded. The signal is output to the pitch adding section 27 as a pitch gain signal.

The spectrum-emphasized audio signal 102 is supplied to a pitch adding unit 27.
And the pitch period signal 112 separately input here.
The audio signal spectrally enhanced for each pitch period represented by the following expression is separated from the end of the frame. An audio signal obtained by multiplying the separated time-domain audio signal by the pitch gain coefficient g in the pitch gain signal is sequentially added to the separated time-domain audio signal. Further, the linear addition audio signal 109 in a state where the audio signal of 20 ms is cut out is output to the temporary memory 24.

Next, the linearly added audio signal stored in the temporary memory is output to the maximum value search correction unit 25, where the maximum value in the frame is searched, and the position and amplitude of the signal on the frame are determined by the multi-pulse signal. Is output to the pulse quantization unit 26 as a part of Further, an autocorrelation signal deformed in accordance with the position and the amplitude (this autocorrelation signal is generated in the maximum value search correction unit 25 by the autocorrelation processing of the impulse response signal 110) is obtained. Furthermore, maximum value search correction unit
At 25, a signal obtained by subtracting the modified autocorrelation signal from the linear addition signal stored in the temporary memory 24 is returned to the temporary memory 24. Here, the maximum value of the signal returned to the temporary memory 24 is searched again, and the position and amplitude of the signal on the frame are output as the second multipulse.

The above-described repetition is continued until the number or amplitude of the multi-pulse of each frame signal reaches a predetermined value, and the multi-pulse signal for each frame is sequentially output to the pulse quantization unit 26.

The amplitude of the multi-pulse signal output to the pulse quantization unit 26 is determined based on the waveform processed by the pitch addition unit 27, for example, the waveform shown in FIG. The amplitude is different from the amplitude of the multi-pulse determined based on the waveform before processing, for example, the waveform shown in FIG.

In FIG. 3, the leftmost portion of T divided by the section length P is not affected by the pitch prediction filter 41 described later.

Therefore, the amplitude of the leftmost multipulse should almost coincide with the amplitude of the multipulse determined based on the waveform before being processed by the pitch adding unit 27. However, as described above, the leftmost waveform is a linearly-added waveform in all sections divided in P units, and their amplitudes do not match.

In this embodiment, this problem is solved by providing the amplitude correction means 6. The amplitude correction means 6 includes a maximum value search unit 61
And an amplitude correction unit 62.

The output waveform of the spectrum emphasis filter 16 is the maximum value search section
Supplied to 61. The maximum value search unit 61 searches for the maximum value in the section indicated by P at the left end in FIG. 3A, and outputs this maximum value to the amplitude correction unit 62. The amplitude correction unit 62 calculates a ratio between the maximum value and the amplitude of the maximum pulse present in the section indicated by P at the left end of the multi-pulses supplied from the pulse quantization unit 26. Further, the amplitude corrector 62 corrects all multi-pulses based on the ratio and outputs the result to the pulse quantizer.

In the pulse quantization unit 26, the input multi-pulse signal is quantized into a quantized multi-pulse signal 111, which is output to the multiplexing / synthesizing unit 35 together with the LPC coefficient signal 104, the pitch period signal 112, and the pitch gain signal 113. You.

The LPC coefficient signal 104, the multi-pulse signal 111, the pitch period signal 112, and the pitch gain signal 113 are input to the multiplexing / combining unit 35, and a multiplexed signal 105 obtained by multiplexing these signals is output from the multiplexing / combining unit 35. You. The multiplexed signal 105 is input to the demultiplexing unit 36 of the waveform decoding unit 4 via a transmission line.

Next, the waveform decoding means 4 will be described. The waveform decoding unit 4 includes a demultiplexing unit 36, a pulse decoding unit 40, a pitch prediction filter 41, a K decoding unit 42, a K · α conversion unit
43, and a pitch gain decoding unit 44. When the multiplexed signal 105 is input from the multiplexing / combining unit 35 of the waveform encoding unit 3 to the multiplexing / demultiplexing unit 36, the multiplexing / demultiplexing unit 36 outputs an LPC coefficient signal 114 corresponding to the LPC coefficient signal 104 and a multi-pulse signal. A multi-pulse signal 121 corresponding to 111, a pitch cycle signal 122 corresponding to the pitch cycle signal 112, and a pitch gain signal
A pitch gain signal 123 corresponding to 113 is output.

The LPC coefficient signal 114 output from the demultiplexing unit 36 is represented by K
The LPC coefficient signal 107 decoded by the decoding unit 42 and further converted by the K / α conversion unit 43 is input to the LPC synthesis unit 5. Similarly, a multipulse signal 121 indicating the position and amplitude of the multipulse output from the demultiplexing unit 36 is decoded by the pulse decoding unit 40 and input to the pitch prediction filter 41. The pitch period signal 122 output from the demultiplexing unit 36 is directly input to the pitch prediction filter 41, and the decoded multipulse signal 106 is input to the LPC synthesis unit 5. The multi-pulse signal 106 is an LPC coefficient signal 107
Are decoded while being controlled in accordance with
Is output from the LPC synthesis unit 5.

〔The invention's effect〕

As described above, the multi-pulse encoding apparatus according to the present invention includes a pitch addition unit that performs linear addition of a speech waveform based on the pitch cycle of the speech waveform, thereby reducing the transmission bandwidth and the number of operations, thereby reducing the number of pulses. The effect is that the search is easy and the structure is simple.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention, FIG. 2 is a block diagram showing an example of a spectrum emphasizing filter constituting the present invention, and FIGS. 3 (a) and 3 (b) show the present invention, respectively. FIGS. 4A and 4B are waveform diagrams of a spectrum-emphasized voice signal showing an operation of pitch addition of FIG. 3A, wherein FIG. 4A is an input voice waveform diagram, and FIG. FIG. DESCRIPTION OF SYMBOLS 1 ... LPC analysis means, 2 ... Spectrum emphasis means, 3 ... Waveform encoding means, 4 ... Waveform decoding means, 5 ... LPC synthesis part, 6
... Amplitude correction means, 11 ... Waveform extraction unit, 12 ... LPC analysis unit, 13
... K quantization decoding section, 14 ... K-α conversion section, 15 ... Attenuation coefficient applying section, 16 ... Spectral emphasis filter, 17 ... Waveform cutout section, 20 ... Impulse response section, 21 ... Spectral emphasis filter, 24 ... Temporary memory, 25: maximum value search / correction unit, 26: pulse quantization unit, 27: pitch addition unit, 31: pitch extraction unit, 32
... Pitch gain quantization decoder, 35 ... Multiplexer, 36 ...
Demultiplexing unit, 40: pulse decoding unit, 41: pitch prediction filter, 42: K decoding unit, 43: K / α conversion unit, 44: pitch gain decoding unit, 61: maximum value search unit, 62 ... Amplitude correction unit.

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G10L 9/14 H03M 1/12 H04B 14/04

Claims (1)

(57) [Claims] 1. A multi-pulse encoder for transmitting, decoding and outputting an encoded signal obtained by encoding an input audio signal, wherein pitch information is extracted from the audio signal, and a pitch period signal and a pitch gain coefficient signal are extracted. And a pitch extracting means for outputting a voice signal in one time range of a time range separated by a pitch cycle from one end of the frame time length by separating the voice signal by a pitch cycle for each frame time length. Linear adding means for sequentially outputting an audio signal to be encoded by sequentially performing operations of adding an audio signal obtained by multiplying an adjacent audio signal in the preceding time range by a pitch gain coefficient to the other end of the frame time length. And an amplitude correction means for correcting the amplitude of the multi-pulse obtained from the audio signal after addition using the audio signal before addition. Luz type encoder.