JPH07154268A - Band division encoder - Google Patents

Abstract

PURPOSE:To suppress quantization distortion caused when the power of a voice signal is normalized by the use of the value of sigma when the power is quantized to the 2sigma-th power of 2. CONSTITUTION:The coder is made up of a band division filter 1 dividing a digital signal such as a voice signal into plural bands, a power calculation means 2 obtaining a power of each block of an output of the filter 1, a power quantization means 3 applying quantization in a form of 2sigma-th power of 2 to the power obtained by the power calculation means 2, a normalizing means 4 normalizing an output of the filter 1 by using the sigma obtained by the power quantization means 3, and a signal encoding means 6 using a quantization table with a corrected error to be estimated when the output of the filter 1 is normalized by using the sigma to quantize an output of the normalizing means 4.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a band division coding apparatus for coding a digital voice signal used in a voice IC recorder or the like.

[0002]

2. Description of the Related Art Conventionally, when a digital audio signal is compression-encoded, a method of dividing the audio signal into several sub-bands and encoding them has been general.

For example, Japanese Patent Application No.
In the specification and drawings of No. 134561, the digital audio signal input in the encoding unit is first divided by a band division filter group (BPF: Band Path Filter), and each of the divided digital audio signals is passed through the filter. The frequency is shifted down by the center frequency of the band, converted there into a low-frequency signal through a low-pass filter, down-sampled at the sampling frequency determined by the bandwidth of the band-pass filter, and the result is forward adaptive for each block. PCM
(Pulse Code Modulation) quantization is disclosed.

According to this proposal, the decoding section performs an operation reverse to the above operation for each band to create a band-pass signal for each band, and adds these to obtain an output speech waveform.

The advantages of such a system are as follows: (1) It is easy to efficiently perform bit allocation for each band. (2) Quantization error in each band relates only to the signal in that band. When the power is low, even if there is a large amount of quantization distortion, it is only a slight error when viewed from the overall signal. (3) The quantization error in each band ideally becomes white noise and spreads widely in the frequency spectrum. It is raised that only a part falls in that band.

From the above advantages (1) and (2), it is desirable to adaptively change the number of bits per sample assigned to each band according to the power of the signal in that band. In this case, It has been said that it is most appropriate to divide according to Equation 1 (the waveform distortion is minimized).

[0007]

[Equation 1]

Here, the electric power obtained for each block is set to 2σ
If quantized to the power of

[0009]

[Equation 2]

[0010] Further, for example, if W _i = 1/4 and N = 4, then the above equation 2 becomes

[0011]

[Equation 3]

[0012] By the way, the distribution of the sample value of the voice waveform or the difference value thereof shows a good approximation to the gamma distribution or the Laplace distribution. Therefore, when quantizing the voice signal for each block described above by the forward adaptive PCM, the optimum quantization table for the gamma distribution or the Laplace distribution may be used. For that purpose, the power of the voice signal becomes 1. Must be normalized as follows.

However, if the power is quantized to the power of 2 2 σ and the value of σ cannot be used for accurate normalization, there is a problem that the quantization distortion increases at this time.

[0014]

When a voice signal is quantized by the forward adaptive PCM as in the above-mentioned prior art, an optimum quantization table for a gamma distribution or a Laplace distribution may be used. The signal must be normalized so that its power is 1.

However, if the power of a voice signal is quantized to the power of 2 2 σ and normalized using σ, it actually becomes larger than 1 and cannot be accurately normalized. At that time, the gamma distribution or the Laplace There is a problem that the quantization distortion increases if an optimum quantization table is used for the distribution. The present invention has been made in view of the problems of the prior art, and suppresses quantization distortion when normalizing using the value of σ when the power of a voice signal is quantized to the power of 2σ. The purpose is to

[0016]

DISCLOSURE OF THE INVENTION The present invention provides a band division filter for dividing a digital signal such as a voice signal into a plurality of bands, a power calculation unit for calculating the power of each block of the output of the filter, and the power calculation. A power quantization means for quantizing the power obtained by the means into a power of 2 2 σ, a normalization means for normalizing the output of the filter using σ obtained by the power quantization means; And a signal encoding means for quantizing the output of the normalizing means using a quantization table that has been subjected to correction of an error expected when the output of the filter is normalized using.

[0017]

As described above, the power of the voice signal is set to 2 2σ
If the quantization table is corrected in advance with the error amount when the power is quantized and the power is normalized using the value of σ, the quantization error when the audio signal is quantized by the forward adaptive PCM is reduced. To do.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the band compression encoding apparatus of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a functional block diagram of a band compression encoding apparatus, in which 1 is a bandpass filter, and an input voice has two stages of QMF (Quadra
(ture mirror filter) and divides it into four bands and drops the output of the band pass filter 1 to the base band.

Reference numeral 2 is a power calculation means for collecting the voices, which have been band-divided for each band as described above, into blocks for every several samples (for example, 128 samples) and calculating the power in the blocks. Reference numeral 3 is a power quantization means for quantizing the power calculated by the power calculation means 2 to the power of 2 to the power of 2 and actually, the calculated power is represented by a binary number, and the value in the represented binary number is represented. Find the place where 1 stands on the leftmost side, find out what bit it is from the right end, subtract 1 from that bit value, and divide by 2 to obtain σ _i for each band Is.

Reference numeral 4 is a normalizing means for normalizing the electric power in one block quantized by the electric power quantizing means 3 to be 1. By this normalizing means 4, it becomes possible to accurately quantize a digital audio signal having a wide dynamic range. Reference numeral 5 denotes a quantizing bit number control unit for allocating the number of quantizing bits used in the signal coding means 6 of each band, which will be described later.

Then, the signal encoding means 6 uses the quantization
According to the number of quantization bits assigned by the bit number control unit 5.
And predict the signal normalized by the normalizing means 4 in advance.
Using a quantization table corrected with the normalized error
It is quantized adaptively. 7 is the signal encoding means 6
Therefore, the code of each band and the auxiliary information σ _i Many
It is a multiplexer that is multiplexed into one signal.

As described above, each of the above-mentioned means 1 to 7 constitutes an audio signal encoding section. On the other hand, the lower part of the decoding of the audio signal is a demultiplexer 8 which divides the signal from the multiplexer 7 into codes and auxiliary information σ _i for each band.
An inverse quantized bit number control unit 9 for calculating bit allocation like the quantized bit number control unit 5, a signal decoding means 10 for decoding the quantized code of each band, and an encoding unit. The denormalization means 11 that restores a voice signal that has been normalized using the value of σ obtained in step 1 above, and the decoding side that up-samples the band-divided and down-sampled signal and band-combines them. It is composed of a bandpass filter 12.

Explaining the operation of the band division encoding device having the above-mentioned structure, first, band division is performed by the band pass filter 1 for any input voice, and the power of each band is calculated by the power calculation means 2. Then, the electric power obtained by the electric power quantizing means 3 is quantized into 2 to the power of 2σ. Then, the quantization bit number control unit 5 adaptively controls the quantization bit number assigned to each band based on the obtained value of σ. By the way, since the power is normalized by the normalizing means 4 based on the value of σ, the output of the normalizing means 4 includes an error. However, a quantization table corrected by an expected normalization error is used. Optimal quantization can be performed by performing quantization by the signal quantization means 6 according to the above.

Further, the quantized signal is decoded into the original digital audio signal by the reverse operation of these operations. Specific examples of the quantization table will be given below. Table 5 to Table 8
Shows an optimum quantization table for the Laplace distribution when the variance of the input audio signal is 1 when quantizing each of 1 bit to 4 bits. The left side of each table shows the absolute value of the step width, and the right side shows the value when reproducing the signal within the range of the step width. Tables 1 to 4 are optimum quantization tables corrected by the previously estimated normalization error when performing the same quantization as in Tables 5 to 8. An error is included in the normalizing means 4 regardless of which one of these tables is used for encoding with 1 bit to 4 bits, and using the corrected quantization table of the present invention, the quantization distortion that minimizes this error is reduced. It can be said that it can be reduced.

[0025]

[Table 1]

[0026]

[Table 2]

[0027]

[Table 3]

[0028]

[Table 4]

[0029]

[Table 5]

[0030]

[Table 6]

[0031]

[Table 7]

[0032]

[Table 8]

[0033]

As described above, according to the present invention, the power for each band is obtained when the digital voice signal is band-division-encoded, and the value is quantized to the power of 2 2σ. This makes it possible to adaptively control the number of quantization bits assigned to each band.

Quantization is performed by adaptively quantizing a signal using a quantization table in which a signal including an error due to normalization based on the value of σ is corrected by an expected normalization error amount. Distortion can be reduced. Therefore, it is possible to expect an effect that it is possible to optimally encode any voice.

[Brief description of drawings]

FIG. 1 is a functional block diagram showing the configuration of an embodiment of a band division encoding device of the present invention.

[Explanation of symbols]

 1, 12 band pass filter 2 power calculation means 3 power quantization means 4 normalization means 5 quantization bit number control means 6 signal coding means 7 multiplexer 8 demultiplexer 9 decoding bit number control means 10 signal decoding means 11 reverse Normalization means

Claims (1)

[Claims] 1. A band division filter for dividing a digital signal such as a voice signal into a plurality of bands, a power calculation unit for calculating the power of each block of the output of the filter, and a power calculation unit for calculating the power calculated by the power calculation unit. Power quantizing means for quantizing to the power of 2σ, normalizing means for normalizing the output of the filter by using σ obtained by the power quantizing means, and output of the filter by using σ A band division encoding device comprising signal encoding means for quantizing the output of the normalizing means using a quantization table which has been subjected to correction of an error expected when normalized.