JPH06208400A - Coding system for code excitation and linear prediction - Google Patents

Abstract

PURPOSE:To provide a coding system for code exciting and linear prediction in which power consumption is less and the deterioration of reproduced sound quality is less. CONSTITUTION:A sound source signal is generated by sum of code books of an adaptive code book 3 storing a sound source signal of the past and a noise code book 6 containing noise signals. When the optimum code book 3 is selected, a SN ratio calculating circuit 11 calculates SN ratio of a sound signal of the frame, and restricts the searching range of the noise code book 6 within a prescribed range when it is judged that this calculated value of SN ratio is larger than the reference value previously decided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a code-excited linear predictive coding system for coding speech.

[0002]

2. Description of the Related Art In a conventional code-excited linear predictive coding system (hereinafter referred to as "CELP"), an adaptive codebook that collects dozens of samples of a speech signal into one vector and accumulates past excitation signals. And a noise codebook prepared in advance, the sum of the respective code vectors is expressed as an output signal that drives the synthesis filter. For each code vector, the optimum one is searched by the A-B-S (analysis by synthesis) method, and the linear prediction coefficient calculated from the index information and the voice signal is transmitted.
Quantization in such vector units is called vector quantization.

FIG. 2 is a block diagram showing an example of a conventional CELP.

In FIG. 2, an input voice signal is input to a subtractor 1, and the output from the synthesis filter 2 is also input to the subtractor 1. The output of the adaptive codebook 3 is given to the synthesis filter 2 via a multiplier 4 and an adder 5. The output of the noise codebook 6 is given to the synthesis filter 2 via a multiplier 7 and the adder 5. The output of the adder 5 is fed back to the adaptive codebook 3.

The subtractor 1 generates an error signal indicating an error between the output of the synthesis filter 2 and the input voice signal and gives it to the auditory weighting filter 8. The perceptual weighting filter 8 gives a signal generated by predetermined perceptual weighting to the error signal to the noise codebook 6 via an error power minimization circuit 9. The optimum lag of the adaptive codebook 3 and the index of the optimum code vector of the noise codebook 6 are sent to the multiplexer 10.

For the adaptive codebook 3, the search for the optimum code vector is carried out over the pitch extraction range Abs.
The noise codebook 6 is searched for in the same manner over the entire range.

[0007]

Therefore, the conventional C
In ELP, the amount of calculation becomes enormous, and in order to finish it within the frame time, high-speed processing is required, and there is a problem that power consumption increases.

An object of the present invention is to provide a CELP which consumes less power and has less deterioration in the quality of reproduced voice.

[0009]

According to the present invention, a sound source is a sum of code vectors of an adaptive codebook accumulating past sound source signals and a noise codebook containing a noise signal prepared in advance. In a code-excited linear predictive coding method for generating a signal, selection determining means for determining that the optimum codebook of the adaptive codebook is selected,
When it is determined by the selection determining means that the optimum codebook is selected, the SN ratio of the audio signal of the corresponding frame is calculated to determine whether or not the calculated SN ratio is larger than a predetermined reference value. SN ratio determination means and this SN
Code-excited linear predictive coding, which comprises range limiting means for limiting the search range of the noise codebook to a predetermined range when the ratio determination means determines that the calculated SN ratio value is larger than the reference value. The scheme is obtained.

[0010]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention. In the embodiment of FIG. 1, the same components as those of the CELP of FIG. 2 are designated by the same reference numerals.

In FIG. 1, the output of the synthesis filter 2 is given to the SN ratio calculation circuit 11. The output of the SN ratio calculation circuit 11 is given to the noise codebook 6.

It is known that when the voice signal is divided into an unvoiced sound region (a noise signal serves as an excitation source) and a voiced sound region (the vocal cord vibration serves as an excitation source), the voiced sound region is larger. . In a voiced frame, the audio signal of the corresponding frame can be almost reproduced with the optimum code vector selected from the adaptive codebook.

In this case, it is considered that the adaptive code vector is dominant in the SN ratio (reproduced voice power to quantization noise power ratio) of the reproduced voice signal, and the noise code vector does not contribute much.

The optimum adaptive code vector is Abs
When selected by the method, the SN ratio calculation circuit 11 calculates the SN ratio with the voice signal of the corresponding frame, and when it is larger than a predetermined threshold value, the search range of the noise codebook 6 is limited (for example, This makes it possible to search the noise codebook 6 in half, for example, in a frame that satisfies the above conditions, and to reduce the power consumption in that frame compared to the full search.

If there are many such frames during the talk time, averaging can be expected to significantly reduce the power consumption of the encoder. Further, since the SN ratio is checked by the optimum adaptive code vector, the reproduced voice is less deteriorated (with respect to the code vector full search of the noise codebook 6).

[0017]

The present invention reduces power consumption, and
It is possible to reduce the deterioration of the quality of reproduced voice.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a code-excited linear predictive coding system according to the present invention.

FIG. 2 is a block diagram showing a conventional code-excited linear predictive coding system.

[Explanation of symbols]

 1 Subtraction circuit 2 Synthesis filter 3 Adaptive codebook 6 Noise codebook 8 Auditory weighting filter 9 Error power minimization circuit

Claims (1)

[Claims] 1. Code-excited linear predictive coding for generating a sound source signal by the sum of code vectors of an adaptive codebook accumulating past sound source signals and a noise codebook containing a noise signal prepared in advance. In the method, selection determining means for determining that the optimum codebook of the adaptive codebook has been selected, and SN ratio of the audio signal of the corresponding frame when the selection determining means determines that the optimum codebook has been selected Is calculated to determine whether the calculated SN ratio is greater than a predetermined reference value, and the SN ratio determination means determines that the calculated SN ratio is greater than the reference value. A code-excited linear predictive coding method, characterized in that the code-excited linear predictive coding method further comprises range limiting means for limiting a search range of the noise codebook to a predetermined range.