JPH0458300A - Vector quantizing method for sound signal - Google Patents

Abstract

PURPOSE:To reduce the processing amount of retrieval as a whole and to reduce energy consumption by defining a relevant code vector as a representative vector when the distortion of a sound to be calculated from an input vector and a code book is lower than an adaptively set threshold value. CONSTITUTION:A discriminating means 1 discriminates whether an input sound is a sound or silence and whether the sound is a vowel or a consonant. A threshold value setting means 2 adaptively changes the threshold value corresponding to the state of the input sound. In order to select the representative vector from a code book 5, a representative vector calculating means 4 successively retrieves code vectors in the code book 5, stops the retrieval when the distorsion of the code vector and of the input vector is made lower than the threshold value, and selects the code vector at that time as the representative vector. Thus, without degrading sound quality by distortion, the operated variable for selecting the representative vector and the energy consumption can be reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to an audio signal vector quantization method for quantizing an A/D converted audio signal.

BACKGROUND ART Conventionally, in this type of vector quantization method, as shown in the flowchart of FIG. 5, first in step S1, there is an input vector X1 X = (X++Xx, . It is input at fixed time intervals (hereinafter referred to as frames).

Next, for the input vector X, a predetermined code vector Yi (i=1.2...N)
, y ix+~13'lI) is calculated for all code vectors Yi, for example, based on the following equation, di=ΣIxi+yLl'' (steps S2.S3.S4°S2).

Next, find i that minimizes the distortion di (step S5)
, the number i is set as the quantized value of the input vector X (step S6).

Problems to be Solved by the Invention However, in the conventional vector quantization method for audio signals described above, when selecting the code vector with the minimum distortion (hereinafter referred to as the representative vector), it is necessary to calculate the distortion of all code vectors. Therefore, there were problems in that the amount of calculation was large and the power consumption was large.

The present invention solves these conventional problems,
The object of the present invention is to provide a vector quantization method that can reduce the amount of calculations and save power consumption.

Means for Solving the Problems In order to achieve the above object, the present invention determines whether an input audio signal is voiced or silent, and if it is voiced, determines whether it is a vowel or a consonant. Then, vector-convert the input audio signal to form an input vector, set a threshold regarding audio distortion calculated from the code vector and the input vector according to the above judgment result, and create a vector consisting of a plurality of code vectors. Refer to the codebook, calculate the audio distortion from the code vector obtained by referring to it, and the input vector,
The calculated distortion is compared with the above threshold value, and if it is above the threshold value, the next code vector is referred to, the above distortion calculation and comparison process with the threshold value is repeated, and if it is below the threshold value, the next code vector is , and selects the code vector as a representative vector, thereby suppressing deterioration of sound quality due to distortion and reducing the overall amount of calculation.

Effects The present invention has the following effects due to the above structure.

In other words, it is determined whether the input voice is a vowel, a consonant, or silent, and if it is silent or a voiceless consonant, there will be little deterioration even if the distortion is large, so the threshold is set to a low value.
For vowels, if the distortion is large, the deterioration is large, so the threshold value is set high. In this way, the threshold value is adaptively changed depending on the state of the input audio. Then, to select a representative vector from the codebook, the codevectors are sequentially searched, and when the speech distortion determined from the codevector in the codebook and the input vector becomes less than a threshold, the search is stopped, and the search is stopped. Determine the code vector as the representative vector. This suppresses deterioration in sound quality due to distortion, and reduces the overall amount of calculation and power consumption due to code vector search.

Embodiment FIG. 1 is a block diagram showing an embodiment of the vector quantization method of the present invention. The embodiment shown in FIG. 1 is a voiced/silent/vowel/consonant determining means l for determining whether an input voice is a vowel or a consonant, and whether it is voiced or silent.
and threshold setting means 2 for changing the threshold according to the result.
, an input vector calculation means 3 for vector-converting input speech, a representative vector calculation means 4 for calculating a representative vector from the input vector and a threshold value, and a codebook 5 storing code vectors used therein.

According to the above embodiment, the voiced/silent/vowel/consonant determining means 1 determines whether the input voice is voiced or silent, and whether it is a vowel or a consonant. When the voiced/silent/vowel/consonant determining means 1 determines that the input voice is silent or a voiceless consonant, the threshold setting means 3 sets the threshold to a low value, since there is little deterioration in sound quality even if the distortion is large. . In addition, when the voiced/silent/vowel/consonant determining means 1 determines that the sound is a vowel, the greater the distortion, the greater the deterioration of the sound quality (1), the threshold setting means 3 sets the threshold to a higher value. do.

That is, the threshold value is adaptively changed depending on the state of the input audio. Then, in order to select a representative vector from the codebook 5, the representative vector calculation means 4 sequentially searches for codevectors in the codebook 5, and stops the search when the distortion between the codevector and the input vector becomes less than a threshold value. , select the code vector at that time as the representative vector. As a result, the amount of calculation and power consumption for selecting a representative vector can be reduced without deteriorating the sound quality due to distortion.

FIG. 2 is a block diagram showing an example of a circuit embodying the embodiment shown in FIG. In Fig. 2, 11 is an A/D converter that converts audio signals into digital signals, 12 is a central processing unit (CPU), and 3 stores an execution program for the CPU 12 and a codebook made up of multiple code vectors. 14 is a random access memory (ROM) in which input vectors are stored.
RAM).

3 is a flowchart showing the operation of the CPU 12 shown in FIG. 2, and FIG. 4 is a flowchart showing the operation of the CPU 12 shown in FIG.
is an explanatory diagram showing the memory contents of the ROM.
13 stores in advance a codebook composed of a plurality of code vectors, and RAM 14 stores input vectors calculated from input speech.

Next, regarding the operation of the circuit shown in FIG. 2, the flowchart shown in FIG. 3 and the ROM 13 and RAM 1 shown in FIG.
This will be explained using an explanatory diagram showing the storage contents of No. 4.

The audio signal converted into a digital signal by the A/D converter 11 is subjected to the following processing for each frame in the CPU 12.

First, the input digital signal is judged in step S12 as to whether or not there is sound in the signal within the frame (silence/sound determination). If it is determined that there is sound, the sound is unvoiced. It is determined whether there is a voiced sound or not (consonant/vowel determination).

Next, in step S13, a threshold value for setting an allowable range of distortion is determined according to the determination result obtained in step S12. If the result obtained in step S12 is, for example, silence or a voiceless consonant, the sound quality will not deteriorate much even if the distortion with the representative vector is large. If it is determined that it is a stationary portion, the sound quality will deteriorate if the distortion is large, so the threshold value is set high so that the difference from the representative vector becomes small.

Next, in step S14, the input digital signal is converted into a form suitable for vector quantization, and the input vector X Input vector X and code vector YiYt = (3' ll+
V 11+ "·+ 3' in) is calculated (step 515). Next, in step S16, it is determined whether the distortion obtained as a result of the calculation is larger or smaller than the threshold value set in step 313. (step 516), if it is less than the set threshold, the search is finished even if it is in the middle of the codebook, and that code vector is determined as the representative vector.If it is more than the set threshold, the next code is selected. Calculate the distortion of the vector (i+1th code vector), and repeat the steps shown in steps S15 and S16 until the distortion becomes less than the threshold.If no codebook with less than the threshold is found until the end of the codebook, search Among the code vectors obtained, the one with the smallest distortion is determined as the representative vector.

As described above, according to the above embodiment, a threshold value is set depending on the nature of the input signal, and if the code vector is less than or equal to the threshold value, the code vector search is completed and it is determined as the representative vector. The amount of calculation and power consumption can be reduced, and deterioration of sound quality due to the difference from the representative vector can be prevented.

Effects of the Invention As is clear from the above embodiments, the present invention adaptively sets a threshold value depending on the nature of the input speech, and when the distortion of the speech obtained from the input vector and the codebook is lower than the adaptively set threshold value. In this method, the search is stopped and the code vector is used as the representative vector, and since the range of allowable distortion varies depending on the nature of the input voice, this reduces the deterioration of sound quality and reduces the overall amount of processing by codebook search. It has the effect of reducing power consumption.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the vector quantization method for audio signals of the present invention, and FIG. 2 is a block diagram showing a circuit configuration in which the embodiment shown in FIG. 1 is implemented using a CPU.
Fig. 3 is a flowchart showing the processing contents of the CPU shown in Fig. 2, Fig. 4 is an explanatory diagram showing an example of the storage contents of the ROM and RAM shown in Fig. 2, and Fig. 5 is a conventional vector quantum of an audio signal. 2 is a flowchart showing a method of DESCRIPTION OF SYMBOLS 1... Voiced/silent/vowel/consonant determination means, 2... Threshold setting means, 3... Input vector calculation means, 4...
Representative vector calculation means, 5... code book, 11.
...A/D converter, 12...Central processing unit (CPU)
, 13...ROM, 14...RAM Agent Patent Attorney Shigetaka Awano One Idiot 1 Figure 2 Figure Figure

Claims (1)

[Claims] Determines whether the input audio signal is voiced or silent, and if it is voiced, determines whether it is a vowel or consonant, and converts the input audio signal into a vector to form an input vector. , according to the above judgment result, set a threshold value for speech distortion calculated from the code vector and the input vector, refer to a codebook consisting of multiple code vectors, and calculate the code vector and input vector obtained by reference. Calculate the distortion of the voice from , compare the calculated distortion with the threshold, and if it is greater than the threshold, refer to the next code vector, repeat the process of calculating the distortion and comparing with the threshold,
If the value is below the threshold, the reference to the next code vector is stopped and that code vector is selected as the representative vector, thereby suppressing the deterioration of sound quality due to distortion and reducing the overall amount of calculation. Vector quantization method.