JPH02183630A - Voice coding system - Google Patents

Abstract

PURPOSE:To suppress the missing of information in a decoded voice signal and the increase in noise due to loop back distortion by adopting a constitution such that it is inhibited when a bit assignment of a band is zero and a prescribed quantity of bit assignment is applied forcedly to the band. CONSTITUTION:An input voice signal is split into plural bands and a band bit assignment calculation section 21 calculates a bit assignment number for each band. The result of calculation is checked and a bit assignment number discrimination section 22 checks whether or not there is any band whose bit assignment number is zero and in the presence of bit assignment zero, a forced bit assignment section 23 assigns a prescribed quantity of bits to the band forcedly. A bit reassignment section 24 applies bit reassignment as to other bands. Thus, even when the voice signal power distribution is largely deviated to a part of the bands, information missing in the decoded voice signal is prevented and the increase of noise due to loop back distortion is suppressed.

Description

[Detailed Description of the Invention] [Features] This invention relates to an audio encoding method that divides an input audio signal into multiple bands and adaptively allocates bits to each band to perform highly efficient encoding such as predictive encoding. .

Bit allocation for a certain band with the aim of suppressing information loss in the reconstructed audio signal and increase in noise due to aliasing distortion, which occurs when the audio signal power distribution is heavily biased toward some bands in the audio encoding method using band division. It is configured to prohibit this when the amount becomes zero and forcibly allocate a predetermined amount of beats/beats to that band.

[Industrial application field]

The present invention relates to an audio encoding method that performs encoding such as predictive encoding by dividing an input audio signal into a plurality of bands and adaptively allocating bits to each band.

[Conventional technology]

As a speech encoding method using the band division described in E,
There is a SBC (Sub Band Coding) encoding system, and one example is APC-A B (Ad
Aptive Predictive Coding
with Adaptive old t allocatto
n: Adaptive bit allocation predictive coding) coding methods are known.

The configuration of an encoder based on the APC-AB method is shown in FIG. The APC-toB method divides the two-person voice into multiple sub-bands using a band-splitting filter and converts each into a low-frequency signal.
Perform adaptive predictive coding on each transform output,
The residual signal between the predicted signal and the input audio signal is encoded for each band. Bits are allocated to each band according to the bias of signal power; in addition, in each band, the residual signal is time-divided and bits are allocated according to the power of each time sub-interval. Le. Therefore, in addition to the encoded signal, there is a transmission path.

The number of allocated bits, prediction coefficients of residual signals, etc. are also sent out as side information.

Predictive encoding is performed using a fixed frame length.

The prediction residual signal within a frame is quantized using the following bit allocation. First, the number of bits Rf(i) allocated to each subband (on the frequency axis) is determined according to the following equation (1).

The adaptive bit allocation number Rt (i, j) according to the intra-frame cross-piece time domain division is determined by first detecting the pitch period, dividing one period period into n equal parts (here, n = 4), and dividing that period into It is determined based on the following equation (2) according to the average power within.

Vf(i) Coco(:, Vf(i) is the band residual power, VL(
i, j) are subinterval residual powers, and Rc is the average number of bits allocated per data sample. Further, 1 is an integer from 1 to 3, j is an integer from 1 to 4, and i=1. 2.
3 correspond to the low, mid, and high subbands, respectively.

j=1, 2, and 3.4 correspond to each time interval in each subband, respectively.

In this way, the average bit allocation number R is calculated according to the predicted residual signal power of each band. In Fig. 1θ, the number of subintervals is 4.
This shows an example of the bit allocation number Rt(i,j) for each partial interval in the case of . Data within a frame is quantized using the number of quantization bits determined from the predictive residual signal power within the subinterval.

[Problem to be solved by the invention]

In the band division encoding method using adaptive bit allocation, as the transmission bit rate decreases, the average bit allocation number R naturally decreases. Particularly when the predicted residual signal power ratio of a band is largely biased between bands9, for example, when the audio signal power is concentrated on the low frequency side and almost absent on the high frequency side.

A case may occur in which the number of bits allocated to the higher band becomes zero.

In this way, if encoding processing is performed with the bit allocation for some of the local areas set to zero, the bit allocation for the 41) area will be forcibly set to zero even though there is actually encoding information for that area. Therefore, significant information loss occurs in the decoded restored audio signal.

The original audio signal cannot be restored well.

FIG. 11 and FIG. 12 are diagrams for explaining such problems. FIG. 11 is a diagram showing the result of spectrum analysis of an input audio signal, and FIG. FIG. 3 is a diagram showing the result of spectrum analysis of a restored audio signal obtained by decoding a signal subjected to 8 kbit/s audio encoding in a state where the allocation is zero. In both figures, the horizontal axis represents time, the vertical axis represents frequency, and the density of the graph represents the magnitude of the audio signal power distribution. As is clear from FIGS. 11 and 12.

There is a noticeable lack of high-frequency audio information in the restored audio signal.

Also, for band division of human voice signals on the encoding side, Q
MF (Quadrature Mirror Fi)
However, when the number of bits allocated to a certain band becomes zero, the aliasing distortion caused by the band splitting filter is not canceled out and appears when each band component is synthesized on the restoration side, resulting in noise. A problem also arises in that the amount increases.

Therefore, an object of the present invention is to prevent the increase in noise due to information loss and aliasing distortion in the reconstructed audio signal caused by the audio signal power distribution being largely biased in some bands in the audio encoding method using band division. shall be.

[Means to solve the problem]

FIG. 1 is a diagram explaining the principle of the present invention. In FIG. 1, 21 is a bandwidth bit allocation calculating section, 22 is a bit allocation number determining section, 23 is a forced bit allocation section, and 24 is a bit re-allocation section.

The audio encoding method according to the present invention is an audio encoding method in which an input audio signal is divided into a plurality of bands, and bits are adaptively allocated to each band for encoding.

When the amount of bit allocation for a certain band becomes zero, this is prohibited and a predetermined amount of bits is forcibly allocated to that band.

[Effect]

The input audio signal is divided into a plurality of bands, and a five-band bit allocation calculation unit 21 calculates the number of bit allocations for each band. The bit allocation number determination unit 22 examines this calculation result and determines whether there is a band in which the number of bits allocated is zero. If there is a band in which the number of bits allocated to The bit re-allocation unit 24 forcibly allocates a predetermined amount of bits to the other bands.

〔Example〕

Three embodiments of the present invention will be described below with reference to the drawings. Second
FIG. 5 is a block diagram showing the configuration of an encoder according to the APCAB audio encoding system as an embodiment of the present invention. Second
In the figure, ] and 2 are band division filters consisting of QMF filters.

The input audio signal X is a low frequency signal XL and a middle frequency signal XM.

and high frequency signal Xn. These low frequency signal XL, middle frequency signal XM, and high frequency signal x11 are input to predictive encoders 3 to 5 and bit allocation section 6, respectively.

The configuration of predictive encoders 3-5 is shown in FIG.

As shown in the figure, the predictive encoder includes a differentiator 10 that generates a predictive residual e from a band audio signal ．．
An inverse quantizer 12 that inversely quantizes the residual code E. Adder 13 that generates a locally decoded signal X' from the dequantized output and the predicted information. It is configured to include a predictor 14 and the like that generate prediction information from the locally decoded signal X.

The configuration of the bit allocation section 6 is shown in FIG. As shown in the figure, the bit allocation section 6 includes a prediction error signal calculation section 61. Prediction residual signal power calculation unit 621 Frequency axis bit allocation unit 631 Time axis bit allocation unit 64. High frequency bit allocation determination unit 65. It is configured to include a high frequency bit allocation section 66°, a low/mid range outside bit allocation section 67, and the like.

The multiplexing unit 7 performs each predetermined ij! The residual codes EL, EM, and EH generated by the I encoders 3 to 5 are multiplexed, converted into transmission codes, and sent to the transmission path.

The operation of the apparatus of this embodiment will be explained below. The input audio signal is converted into a low frequency signal XL by band division filters 1 and 2.

It is divided into a mid-range signal XM and a high-range signal X■.

The signals are input to predictive encoders 3 to 5, respectively. Predictive encoder 3
In steps 5 to 5, predictive coding is performed for each band.

Residual codes EL, EM, and EH are obtained and sent to the multiplexing section 7, respectively.

On the other hand, signals XL, XM, and Xn of each band are transmitted to the bit allocation section 6
Here, the number of focus assignments Rf(
i) and the number of bit allocations R for each time subinterval in each band
t(i,j) are calculated and sent to predictive encoders 3 to 5, respectively.
will be sent to.

The bit allocation processing procedure in this bit allocation section 6 is shown in the flowchart of FIG. First, in the bit allocation section 6, each input band signal XL is inputted.

Based on Xv,
) and subinterval residual power V L (i, j) are determined.

The frequency axis bit allocation unit 63 calculates the number of bit allocations Rf(i) on the frequency axis based on the obtained band residual power V f (i) and subinterval residual power according to the above-mentioned formula (1). Calculations are made for each of the low (Hibiki = 1), mid (i = 2), and high (i = 3) bands (step Sl
).

Next, the time axis bit allocation unit 64 calculates the number of bit allocations Rt (i, j) on the time axis according to the above-mentioned equation (2) based on the determined bit allocation number Rf(υ).

Find the time subinterval of each band (step S2)
, calculate the average focus allocation number Rd per data l sample based on this bit allocation number Rt (1, J) (Step S3), and calculate the correction coefficient D1 by comparing this bit allocation number Rd with Rc. (Step S
4).

That is, Re and Rd are compared, and when Rc≧Rd, the correction number D is increased, and when Rc<Rd, the correction number Dr is decreased (step S4).This time axis bit allocation process is repeated m times. (Steps 82 to S5), the number of allocated bits Rt(i,j) for each time sub-interval for each band is determined.

Next, the high frequency bit allocation determining unit 65 determines whether or not the number of bit allocations Rt(3,j) for the time sub-interval in the high frequency band among the bit allocation numbers obtained as described above are all "0". If the bit allocation number Rt (i, j) is not 0 (step 56), the already determined bit allocation number Rt (i, j) is used as is and the process ends.

If the bit allocation number Rt (3, 0 is O), the bit allocation is redistributed, and first the high frequency bit allocation unit 6
6, the subinterval residual power of the high band Vt(3,j)
Step S7) extracts the partial section with the maximum value VL(i, m) from among them, and calculates the number of focus assignments Rt(3) for that partial section.
,...) is set to "1" (step 38).

That is, as shown in FIG. 6, the magnitudes of the high-frequency subinterval prediction error signal powers are compared, and one bit is assigned to the subinterval having the maximum power. In this Figure 6, the horizontal axis is the partial section number.

The general axis is the number of quantization bits, and the 3 subinterval number “1”
” is assigned one bit as the maximum power section.

Next, in the low/mid range bit re-allocation section 67, the number of bits allocated for the high range Rt(3,j) is set as a fixed value, and
=1) and only for the midrange (i=2).

Based on equation (2) above, the number of bits allocated on the time axis Rt(
i, j) while calculating the average bit allocation number Rd and correction number Dr (steps SIO and 5ll),
This is repeated m times (steps 89 to 512). As a result, the reallocated bit allocation number Rt (+, j) takes into account the increased bit allocation ζ number Rt (3, j) in the high frequency range.
and Rt(2,j) are obtained.

Based on the number of time subinterval bit allocations RL (i, j) for each band obtained as described above, each prediction code 5,
j to 5 are residual codes EL and EM obtained by determining the number of quantization bits and performing encoding.

Eu is sent to the multiplexing section 7, where it is multiplexed and sent out to the transmission path.

Various modifications are possible in implementing the invention. For example, in the above embodiment, as a forced bit allocation process in the high frequency range, one bit was allocated to the partial interval having the maximum power as shown in FIG. 6, but the invention is not limited to this. , as shown in FIG. 7, the magnitude of the high-frequency subinterval predicted residual signal power is determined, and one bit is assigned to each of the subinterval having the maximum power and the subinterval having the next largest predicted residual signal power. Alternatively, as shown in FIG. 8, the magnitude of the high-frequency sub-interval prediction residual signal power is determined and one bit is assigned every l pitch period to the sub-interval having the maximum power. You can do it like this.

Further, in the above-described embodiment, a case has been described in which forced bit allocation is performed when the number of bits allocated to the high band becomes 0, but the invention is not limited to this.

Alternatively, forced bit allocation may be performed even when the number of bits allocated in the middle range becomes 0.

Furthermore, the audio encoding method to which the present invention is applied is as follows.

It is not limited to the above-mentioned APC-AB method, but also other SBC encoding methods or A ``r' C (Adaptive Tr
Of course, it is not possible to use a coding method (transform coding) or the like.

〔Effect of the invention〕

According to the present invention, even when the audio signal power distribution is largely biased toward some bands in the audio encoding method using band division, information loss in the reconstructed audio signal can be prevented, and an increase in noise due to aliasing distortion can be prevented. ttn control.

[Brief explanation of the drawing]

FIG. 1 is a diagram explaining the principle of the present invention. FIG. 2 is a block diagram showing an example of the configuration of an encoder using a speech encoding method as an embodiment of the present invention. FIG. 3 is a block diagram showing a configuration example of a predictive encoder in the encoder of the embodiment. FIG. 4 is a block diagram showing an example of the configuration of a bit allocation section in the encoder of the embodiment. FIG. 5 is a flowchart showing the procedure of bit allocation processing in the bit allocation section. FIG. 6 is a diagram showing an example of a forced bit allocation method in the bit allocation section. FIGS. 7 and 8 respectively show forced bit allocation in the bit allocation section as 463 embodiments of the present invention A diagram showing the method. Fig. 9 is a block diagram showing an encoder of a conventional band division type audio encoding system. Fig. 10 is a diagram showing an example of partial interval division of the band division type audio encoding system. 12 is a diagram showing a spectrum analysis result of an input audio signal, and FIG. 12 is a diagram showing a spectrum analysis result of a restored signal of an audio signal subjected to 8 bps encoding processing. In the figures. 1, 2 Band division Filters 3 to 5 - Prediction encoder 6 - Bit allocation unit 10 - Differential unit 11 - Quantizer 12 - Inverse quantizer 13 - Adder 14 - Predictor 61 Prediction residual signal calculation unit 62 - Child side residual Signal power calculation section 63 - Frequency axis bit allocation section 64 - Time axis bit allocation section 65 - High frequency bit allocation determination section Main house - 2 system S load theory device Fig. 1 Fig. 3 The scope of the present invention Fig. 2 Pill - 蓚 [Tongue part 謬駈轲 Wash @ Figure 4 - Door damage, l 唖乃 50 Figure 5 The ridge is j bi7. ¥2 and D, i Figure 8 Figure 9 Partial division between parts Figure 10''; 1221 Procedural amendment (method) May 15, 1999 1999 Patent Application No. 3305 2 Invention Name of the person making the audio encoding method 3 correction Relationship with the μ case Patent applicant address 1015 Kamiodanaka, Nakahara-ku, Yozaki City, Kanagawa Prefecture Name (522) Fujitsu Limited 4th Representative Director Address 211 (044)
754-3035 Address Nakahara-ku, Yozaki-shi, Kanagawa Prefecture 10!5 5 Date of amendment order: March 31, 1999 (Despatch: April 25, 1999) 6 Subject of amendment (1 ) “Detailed Description of the Invention” column in the specification (2) “Brief Description of Drawings” column in the specification (3) Contents of amendment to drawing 7 (1) From page 5, line 14 of the specification By the 5th line of page 6, there is a statement that says, ``Figures 11 and 12 occur...conspicuously.''[For example, some people... The result of spectral analysis of the input audio signal is 8.
A comparison of the results of spectrum analysis of the restored audio signal obtained by restoring a signal subjected to kbit/s audio encoding revealed a phenomenon in which high-frequency audio information is noticeably missing in the restored audio signal. It was observed. ” he corrected. (2) From page 15, line 15 to line 20 of the specification, the statement ``Figure 10 is a diagram showing...'' has been replaced with [
FIG. 10 is a diagram illustrating an example of partial interval division in the band division type speech encoding system. ” he corrected. (3) Figures 11 and 12 of the drawings will be deleted. that's all

Claims (1)

[Claims] In an audio encoding method that divides an input audio signal into a plurality of bands and adaptively allocates bits to each band for encoding, when the amount of bit allocation for a certain band becomes zero, An audio encoding method configured to forcibly allocate a predetermined amount of bits to that band.