Band dividing and coding device

Description: TECHNICAL FIELD [Detailed description of the invention]

【0001】

[Industrial application field]

The present invention relates to a band division coding device that mainly encodes a digital voice signal used in a voice IC recorder or the like.

【0002】

[Conventional technology]

Conventionally, when compressing and coding a digital audio signal, a method of dividing into several sub-bands and coding has been adopted.

【0003】

For example, Japanese Patent Application Laid-Open No. 4-129430 (H03M 7/00) by the applicant of the present invention discloses an example thereof. According to the publication, the input digital audio signal is first divided by a band path filter (BPF), and each of the divided digital audio signals is frequency-shifted down by the central frequency of the pass band. Then, it is converted into a low frequency signal through a low pass filter, downsampled at the sampling frequency determined by the bandwidth of the band pass filter, and the result is encoded by the forward adaptive PCM (Pulse Code Modulation) for each block. It was a thing.

【0004】

Further, in the decoding section of this publication, a band-passing signal for each band is created by performing an operation opposite to the above operation for each band, and these are added together to obtain an output voice waveform.

【0005】

The advantages of such technology are (1) it is easy to effectively allocate bits for each band, (2) the quantization error of each band is related only to the signal of that band, and the power of that band. If is small, even if there is a lot of quantization distortion, there is only a slight error when viewed from the entire signal. (3) The quantization error in each band is ideally white noise and spreads widely in the frequency spectrum. For example, only the part falls into that band.

【0006】

Also, from (1) and (2) above, it is desirable to adaptively change the number of bits per sample allocated to each band according to the power of the signal in that band. In that case, divide according to the following equation. Was considered to be the most appropriate (minimize waveform distortion).

【0007】

[Number 1]

<img file="JPH06348294A_D0001.tif" />【0008】

Here, if the power obtained for each block is quantized in the form of 2 to the power of 2σ, the above equation 1 will be [0009]

[Number 2]

<img file="JPH06348294A_D0002.tif" />【0010】

Will be. Also, for example, in the above number 2, W_i If = 1/4 and N = 4, [0011]

[Number 3]

<img file="JPH06348294A_D0003.tif" />【0012】

Will be.

【0013】

By the way, in this case R_i Can be calculated as a decimal number or a negative number, so a conditional statement was needed to make those numbers positive integers (including 0). And the number of bits R to allocate to each band_i When is a decimal number, it must be rounded to a positive integer, but at that time, there is a problem that the waveform distortion becomes large.

【0014】

Also R_i When is a negative number, the waveform distortion can be minimized by setting the number of bits to be assigned to 0. However, if the number of blocks to which 0 bits are assigned increases, there is a risk that abnormal noise will occur in the reproduced sound. Therefore, in order to prevent this abnormal noise, it is necessary to allocate one or more bits to each band, and at that time, there is a drawback that the number of quantization bits is insufficient as a whole and the quantization error increases.

【0015】

[Problems to be Solved by the Invention]

Number of bits allocated to each band when waveform distortion is minimized as in the conventional technique R_i Can be found in the above number 1. But R_i When is obtained as a decimal number, there is a problem that a conditional statement to make it a positive integer by rounding is required, and in that case, the waveform distortion becomes large.

【0016】

Also R_i When becomes a negative number, the waveform distortion can be minimized by setting the number of bits to be assigned to 0. However, if the number of blocks to which 0 bits are assigned increases, abnormal noise will occur in the reproduced sound. Therefore, in order to prevent this abnormal noise, it is sufficient to allocate 1 bit or more to the band, but at that time, there is a problem that the quantization error increases.

【0017】

The present invention has been made to solve the conventional problem that the waveform distortion becomes large and the problem that the quantization error increases.

【0018】

[Means for solving problems]

The present invention includes a band division filter that divides a digital signal such as an audio signal into a plurality of bands, a power calculation means that obtains the power of each block of the output of the filter and quantizes it in the form of 2 to the power of 2σ, and the block. It is provided with a quantization means for adaptive quantization for each and a quantization bit number control unit for adaptively controlling the number of quantization bits for each band of the quantization means by the value of σ, and the number of quantization bits. The control unit regularly changes the number of quantization bits for each sample data output from the band division filter, and sets the number of quantization bits of the quantization means assigned to each band to a number larger than 0 including a fraction. The quantization bit number control unit preferably sets the number of bits larger than 0 and less than 1 for the band in which the number of bits allocated to the quantization means is 0 or less.

【0019】

[Action]

In the above configuration, the number of bits R allocated to each band when the waveform distortion is minimized as in the prior art._i Is obtained by the above equation 1. And R_i When is a decimal number, R is the number of bits allocated to each band by the above means._i It can be set to a value that matches or is close to, and waveform distortion can be reduced.

【0020】

Also R_i When becomes a negative number, allophones are removed by allocating at least more than 0 bits less than 1 to any band without increasing the quantization distortion of the other bands. Can be done.

【0021】

[Example]

Hereinafter, an embodiment of the present invention will be described in detail based on the functional block diagram of FIG. 1 and the bit number allocation explanatory diagram of FIGS. 2 and 3.

【0022】

In FIG. 1, 1 is a QMF (Qudrature Mirror Filter), which divides the input sound into high and low frequencies at a frequency of 1/4 of the sampling frequency and downsamples to 1/2 of the sampling frequency. Then, by using QMF1 in two stages, the input sound is divided into four bands, and the output of the filter 1 is dropped to the baseband.

【0023】

2 is a power calculation means that collects the voice band-divided for each band into blocks for every few samples (for example, every 128 samples) and calculates the power in the block. 3 is a quantization means that quantizes the calculated power to the 2σth power of 2, and actually represents the power in binary, searches for the place where 1 stands on the far left, and that is the number of bits from the right. Σ for each band by checking if there is, subtracting 1 from that value and dividing by 2._i To ask.

【0024】

4 is a normalization means that normalizes the power in one block so that it becomes 1, and it is possible to quantize a digital audio signal with a wide dynamic range with high accuracy.

【0025】

Reference numeral 5 denotes a quantization bit number control unit that allocates the number of quantization bits used in the quantizer of each band, and can be assigned by the equation 4 using the value of σ of each band.

【0026】

[Number 4]

<img file="JPH06348294A_D0004.tif" />【0027】

Actually R_i May be calculated as a decimal number greater than 0, and R of the above number 4_i There are four values that can be taken by x.00, x.25, x.50, x.75 (x is an integer). At this time, as shown in 1 (1) to (4) of FIG. 2, the average number of bits allocated to each band is reduced to a decimal by regularly changing the number of bits allocated to every other sample.

【0028】

That is, in (1), all samples are quantized with x bits, in (2), 3 out of 4 samples are quantized with x bits, and the remaining 1 sample is quantized with x + 1 bits, (3). In (4), 2 of the 4 samples are quantized with x bits and the remaining 2 samples are quantized with x + 1 bits. In (4), 1 of the 4 samples is quantized with x bits and the remaining 3 samples are x +. The desired fractional bit can be obtained by quantization with 1 bit.

【0029】

Also R_i May be calculated as 0 or a negative number, in which case the number of bits greater than 0 and less than 1 is assigned as shown in Fig. 3. That is, in (1), 1 sample out of 4 samples is quantized with 1 bit and the remaining 3 samples are set to 0. In (2), 2 samples out of 4 samples are quantized with 1 bit and the remaining 2 samples are set to 0. In (3), 3 out of 4 samples are quantized with 1 bit, and the remaining 1 sample is set to 0 to obtain a desired decimal bit.

【0030】

6 is a signal quantization means that quantizes the signal normalized by the normalization means 4 according to the number of quantization bits assigned by the quantization bit number control unit 5, and 7 is a quantized code of each band. And auxiliary information σ_i It is a multiplexer that multiplexes.

【0031】

8 is the signal from the multiplexer 7 with the code and auxiliary information σ of each band._i It is a demultiplexer that divides into.

【0032】

9 is a decoding bit number control unit that calculates the bit allocation in the same manner as the quantization bit number control unit 5, 10 is a power decoding means that decodes the power of each band, and 11 is a power decoding means that decodes the power of each band. Power decoding means, 12 is a denormalization means for restoring a signal that has been normalized using the decoded power.

【0033】

Reference numeral 13 denotes a QMF on the decoding side, which upsamples and band-couples the band-divided and down-sampled signal.

【0034】

With the above configuration, for any voice, the filter 1 divides the band to obtain the power of each band, and the quantization means 6 quantizes the value to the 2σth power of 2 to obtain the value of σ. Based on this, the number of quantization bits allocated to each band can be adaptively controlled by the quantization bit number control unit 5 with the number of bits including fractions, so that optimum coding can be performed.

【0035】

Further, by following the operation opposite to the previous operation, the encoded signal can be decoded into the original digital audio signal.

【0036】

[Effect of the invention]

According to the apparatus of the present invention, when a digital audio signal is band-divided coded, the power for each band is obtained, and the value is quantized in the form of 2 to the power of 2σ to allocate the number of quantization bits to each band. It is possible to adaptively control the number of bits, and by regularly changing the number of quantization bits every other sample, it is possible to cope with the fact that the number of bits allocated to each band becomes a small number. Increase the quantization distortion of other bands by allocating at least the number of bits greater than 0 and less than 1 to the band where the number of allocated bits is 0 or less to reduce the quantization distortion and further minimize the waveform distortion. Since it is possible to remove audible abnormal sounds without causing them to be generated, it is extremely useful because it can be optimally encoded for any voice.

[Simple explanation of drawings]

[Figure 1]

FIG. 1 is a block diagram for explaining an embodiment of the apparatus of the present invention.

[Figure 2]

It is a figure for demonstrating the decimal bit allocation.

[Fig. 3]

It is a figure for demonstrating the allocation of the number of bits larger than 0 and less than 1.

[Explanation of symbols]

1 Bandwidth filter (QMF) 2 Electric power calculation means 5 Quantization bit number control unit 6 Quantization means