JP3016313B2 - Multi-pulse speech encoder - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a multi-pulse type coding apparatus, and more particularly to a multi-pulse coding apparatus for coding by assigning quantization bits for the generation time and amplitude of a multi-pulse.

[0002]

2. Description of the Related Art An input audio signal is analyzed, spectrum envelope information and sound source information constituting the input audio signal information are extracted on an analysis side, and these audio information are transferred to a synthesis side via a transmission path. A multi-pulse speech encoding / decoding device for synthesizing speech has recently been well known.

In a multi-pulse speech coding apparatus,
Generally, the spectral envelope information represents spectral distribution information of a vocal tract system that generates an input voice signal, and is a number of LPC coefficients α corresponding to the analysis order obtained by linear predictive coding (LPC) analysis. Generally, parameters and K parameters are used.

On the other hand, the sound source information indicates the fine structure of the spectrum and is called a residual signal obtained by removing the spectrum envelope information from the input speech signal. The pulse signal is composed of a plurality of impulses having a degree of freedom in the amplitude position. Multi-pulse).

[0005] Since the above-mentioned spectral envelope is obtained with a predetermined number of bits, it is assigned to predetermined quantized bits without special rules. On the other hand, the multi-pulse represented by the amplitude and the position has an amplitude of 0 for compensating the pulse interval by deleting one pulse in the multi-pulse when there is a pulse interval exceeding a limited interval of a preset length. A (zero) pulse is set and quantization is performed. The above operation is repeated to assign the amplitude and position of the multi-pulse to quantization bits and quantize the multi-pulse train.

On the synthesizing side of the LPC coefficient and the multi-pulse train, a synthesized voice is generated as a synthesis filter coefficient and sound source information.

A pulse interval encoding apparatus provided with such a multipulse pulse interval limiting means is disclosed in Japanese Patent Application Laid-Open No. 60-26 / 1985.
No. 0099, “Multi-pulse encoding / decoding device”.

[0008]

However, the conventional multi-pulse speech codec has the following drawbacks.

FIG. 3 is a diagram in which a pulse train is represented by an amplitude and an interval, and the pulse name is P _{m + x} and the interval between adjacent pulses is L _{m + x} (x is..., -2, -1, 0). , + 1, +
2, ...).

FIG. 4 is a diagram in which the pulse train of FIG. 3 is interval-coded by a conventional technique.

[0011] In the prior art, a pulse (hereinafter the amplitude 0 (zero) to assign a pulse interval L _m-3 of that transcends quantization bits L _Q of the preset interval of quantization bits, referred to as dummy pulses Pd )). Therefore, the pulse P _m having the minimum amplitude is deleted, and the intervals (L _m and L _{m + 1} ) adjacent to the deleted pulse are updated. In the drawings, the interval L _m and L _{m + 1} adjacent is updated to the distance L _{m + 1} 'of the adder 4 in Figure 3 pulses P _m.

However, the updated interval (L in FIG. 4)
_{If (m + 1} ) exceeds the limited pulse interval (L _Q ), a dummy pulse is set again, and the pulse with the minimum amplitude is deleted, which greatly consumes the number of quantization bits in the interval. There are drawbacks.

In the multi-pulse speech encoding / decoding device, deleting the multi-pulse results in a decrease in speech information, which leads to deterioration in sound quality of output speech on the synthesis side.

[0014]

According to the present invention, an input speech signal is subjected to LPC analysis for each analysis frame and extracted and LP
The C coefficient is defined as spectrum envelope information, and the sound source information constituting the speech information of the input speech signal together with the spectrum envelope information is a plurality of predetermined sound source information having a generation time position and an amplitude corresponding to the feature of the sound source information for each analysis frame. Expressed by a multipulse that is a sequence of impulse,
In the multi-pulse speech encoding device for analyzing the input signal, in the quantization time allocation of the multi-pulse generation time and amplitude, if there is a pulse interval exceeding a preset pulse interval, set a dummy pulse, When deleting a pulse with a small amplitude, it has interval update inspection means to inspect the interval of the pulse to be deleted, and selects the pulse to be deleted under the condition that the pulse interval derived by deleting the pulse does not exceed the previous pulse interval Is obtained.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the present invention will be described with reference to the drawings. FIG. 1 is a block diagram of one embodiment of the present invention. FIG. 2 is a diagram showing a pulse train after quantizing and encoding the pulse train shown in FIG. 3 in the present invention.

The multi-pulse speech encoding / decoding apparatus of the embodiment shown in FIG. 1 is composed of an analysis side A and a synthesis side B. On the analysis side A, the input speech waveform 10 cut out by the analysis frame length is supplied to the waveform memory 100, and the input speech waveform 11 outputted therefrom is converted into the LPC analyzer 10
1 and the multi-pulse analyzer 103.

The LPC analyzer 101 performs a linear predictive coding analysis on the basis of the input speech waveform 11 and extracts LPC coefficients 12 of a preset order. In this way LPC
The LPC coefficient 12 output from the analyzer 101 is
It is supplied to a coefficient quantizer 102 and a multi-pulse analyzer 103.

The multi-pulse analyzer 103 inputs the input speech waveform 11 and the LPC coefficient 12, and performs multi-pulse analysis based on correlation processing. Then, the amplitude and position of the multi-pulse are determined. This multi-pulse train 15 is supplied to the pulse temporary storage 104 and stored therein.

The multi-pulse train (FIG. 3) 16 output from the temporary pulse storage 104 is input to both the pulse interval check counter 105 and the minimum amplitude searcher 106.

The pulse interval inspection counter 105 clears the count at the pulse position, and counts during a period in which no pulse exists. When the count advances to a preset count number (the maximum value of the pulse interval quantization bit number), an interval update signal 19 is output to the interval update inspector 107. This interval update signal 19 is a signal indicating that dummy pulse generation is necessary.

On the other hand, the minimum amplitude searcher 106 searches the position of the minimum pulse amplitude from the multi-pulse train 16 output from the temporary pulse storage 104, and the interval update inspector 1
07. Here, the minimum amplitude searcher 106 does not search the amplitude position once searched in the same frame again. Further, immediately after outputting the minimum pulse amplitude position 21, the search for the next minimum pulse amplitude position is performed.

Generating and inserting one dummy pulse means deleting one pulse from the pulse train. In the present application, the interval update inspector 107 searches for a pulse to be deleted. Hereinafter, the operation of the interval update inspector 107 will be described. The interval update checker 107 outputs the interval update signal 1
When 9 is input, it is first determined whether or not the minimum amplitude pulse input from the minimum amplitude searcher 106 should be deleted. More specifically, the interval between adjacent amplitude pulses (L _{m + 1 in} FIG. 3) of the interval between minimum amplitude positions (L _{m in} FIG. 3) input from the minimum amplitude searcher 106 is added to calculate the interval quantization bit number. Compared with the maximum value (L _Q ), when Expression (1) is satisfied, the minimum amplitude pulse is deleted.

L _{m +} L _{(m + 1)} <L _Q Equation (1) At this time, the interval update inspector 107 outputs the pulse deletion signal 22.

On the other hand, if the comparison result does not satisfy the equation (1), it is determined that this minimum amplitude pulse should not be deleted. Position of minimum pulse amplitude (P _{m + 2 in} FIG. 3) 2
1 is input again, and it is determined whether this should be deleted. Also in this case, adjacent pulse intervals (L _{m + 2} and L
_{m + 3} ) is added, and it is determined whether Expression (1) is satisfied.

In the case of the multi-pulse train shown in FIG. 3, the sum of the pulse intervals before and after the minimum amplitude pulse Pm is (L _m + L
than _{(m + 1))>} L _Q, the pulse P _m without being satisfy the equation (1) is determined not to be deleted, for the minimum pulse P _{m + 2} of the second _{2, (L m + 2 +} L m ₊₃ ) <L _Q satisfies equation (1), and it is determined that pulse P _{m + 2} should be deleted.

In the present invention, the above operation is repeated to search for a pulse position satisfying the expression (1).

When the pulse deletion signal 22 is input, the dummy pulse generator 108 deletes the pulse at the pulse position satisfying the expression (1), and places the dummy pulse at the position where the pulse interval check counter 105 outputs the interval update signal 19. , A memory change signal 20 is output to rewrite the contents of the pulse temporary storage. The pulse temporary storage 104
In response to the memory change signal 20, the dummy pulse which does not affect the amplitude is stored, and the pulse to be deleted is deleted from the storage.

After outputting the interval update signal 19, the pulse interval inspection counter 105 decrements the count value.
The inspection count starts from the dummy pulse position.

In the present invention, the interval update inspector 107
Is used to delete the pulse associated with the setting of the dummy pulse, the interval between the positions where the pulse is deleted (L _{m + 3} ′ in FIG. 2)
But, it does not exceed the maximum value L _Q of the interval quantization. For this reason, the number of deleted pulses is smaller than that in the related art.

According to the result of an experiment using a real voice sample for about 1 minute, a multi-pulse type voice coding / decoding apparatus in which the coding speed is 9600 BPS, the frame period is 20 mSec, and the number of searched multi-pulses is 19 In contrast, the number of pulses after encoding was 15.4 pulses on average in the prior art, while the average number of pulses was 16.4 when using the method of the present invention.
Five pulses are obtained, which is improved by 1.1 pulses per frame. This is 3 bits quantization amplitude per pulse, 5 intervals
About 440 BPS (ビ ッ ト (16.5-15.
4) × (3 + 5) / 20 mSec) This corresponds to an improvement in coding efficiency.

When the multi-pulse train for one frame is inspected and counted, the contents of the pulse temporary storage 104 are output to the multiplexer 106. The multipulse 109 multiplexes the LPC coefficient 14 and the multipulse train 17 and outputs multipulse information 23 to the transmission path.

On the other hand, on the synthesizing side B of the multi-pulse encoding / decoding device, the demultiplexer 110 receives these multiplexed data via a transmission line and separates the LPC coefficient 24 and the multi-pulse train 26. .

The LPC coefficient 24 is supplied to the LPC decoder 111, and the multi-pulse train 26 is supplied to the pulse decoder 112. LPC decoder 111 is:
The LPC coefficient 24 is input and subjected to decoding processing, and the LPC coefficient 25 is output.

The multi-pulse decoder 112 receives the multi-pulse train 26, performs a decoding process, and outputs a multi-pulse train 27. The LPC synthesis filter 113 is an LPC synthesis filter.
A speech is synthesized and output using the LPC coefficient 25 input from the decoder 111 as a filter coefficient and the multi-pulse train 27 input from the pulse decoder 112 as sound source information.

[0035]

As described above, the present invention reduces the number of pulses of the multi-pulse train which is the excitation information of the multi-pulse encoder by reducing the consumption of the number of quantization bits of the pulse interval, thereby reducing the number of pulses of the synthesized speech. There is an effect that the sound quality can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a circuit configuration of an embodiment of the present invention.

FIG. 2 is a diagram showing a pulse train obtained by encoding the multi-pulse train of FIG. 3 according to the present invention.

FIG. 3 is a diagram showing a multi-pulse train analyzed by a multi-pulse analyzer.

FIG. 4 is a diagram showing a pulse train obtained by encoding the multi-pulse train of FIG. 3 by a conventional technique.

[Explanation of symbols]

 REFERENCE SIGNS LIST 100 Waveform memory 101 LPC analyzer 102 LPC coefficient quantizer 103 Multi-pulse analyzer 104 Pulse temporary storage 105 Pulse interval check counter 106 Minimum amplitude searcher 107 Interval update inspector 108 Dummy pulse generator 111 LPC decoder 112 pulse Decoder 113 LPC synthesis filter

Claims (2)

(57) [Claims] An input audio signal is converted to an LP for each analysis frame.
The LPC coefficients extracted by C analysis are used as spectral envelope information, and the sound source information constituting the sound information of the input sound signal is a predetermined sound source having an occurrence time position and an amplitude corresponding to the feature of the sound source information for each analysis frame. In a multi-pulse speech encoding device that encodes the input signal by expressing the input signal as a multi-pulse sequence of multi-pulses, a pulse interval exceeding a preset pulse interval exists in the multi-pulse train. In this case, a multi-pulse quantizer that sets a dummy pulse during this period and deletes a pulse having a relatively small amplitude to quantize the multi-pulse train is included.In the multi-pulse quantizer, a pulse is deleted. Select pulses to be deleted on condition that the pulse interval to be created does not exceed the preset pulse interval. Multi-pulse-type speech coding apparatus according to claim. 2. The multi-pulse quantizer according to claim 1, wherein said multi-pulse quantizer measures a pulse interval of the multi-pulse read from said storage means, and stores said pulse sequence when said pulse interval exceeds a preset pulse interval. A measuring unit that outputs an output to generate a dummy pulse, a search unit that searches for a multipulse read from the storage unit, and a search unit that prepares a search result in an order of small amplitude; and when an output is output from the measuring unit, 2. The multi-function device according to claim 1, further comprising: a determination unit configured to determine whether a sum of pulse intervals before and after a pulse having a smaller amplitude prepared by the search unit exceeds the preset pulse interval. Pulse type speech coding device.