Process for compressing data relative to voice signals and device applying said process

• US 4,216,354 A
• Filed: 11/29/1978
• Issued: 08/05/1980
• Est. Priority Date: 12/23/1977
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Mauduit, Daniel, Menez, Jean, Esteban, Daniel J., Galand, Claude R.
• Primary Examiner(s): Morrison, Malcolm A.
• Assistant Examiner(s): Kemeny, E. S.
• Application Number:
  US05/964,324
• Time in Patent Office: 615 Days
• Field of Search: 179/15.55 R, 179/15.55 T, 179/1.5 A
• US Class Current: 704/229
• CPC Class Codes: H04B 1/667 using a division in frequency subbands for TV signals H04N19/63

Abstract

A voice signal is transmitted digitally at reduced bit rate by use of data compression. The original frequency components of a telephone band width of the voice signal sampled at 8 KHz and quantized with 12 bits are transformed into three parameters: SIGNAL data representing adaptive quantization for lower frequency (300-800 Hz) sub-bands; COEF data representing pre-emphasized parcor type coefficients for the higher frequency (800-3000 Hz) band; and, ENERGY data representing higher frequency short term energy level. The three parameters are multiplexed for transmission in binary-code form, thereby representing a recoding of the original binary-coded voice signals.

Claims

1. Process for compressing a digitally encoded voice signal, said compression resulting from a transcoding operation, characterized in that it includes the following steps:

on a recuring basis splitting an epoch of the frequency bandwidth of said voice signal into a low frequency and a high frequency band, digitally coding and requantizing with dynamic bit allocation for each epoch the contents of said low frequency band for each epoch to generate SIGNAL information, processing the contents of said high frequency band for each epoch to generate a set of preemphasised parcor type coefficients (COEF), processing the contents of said high frequency band for each epoch to generate energy information (ENERG), said SIGNAL, COEF and ENERG data generated for each epoch providing the digital data characterizing the compressed voice signal.

2. Process for digitally compressing a voice signal according to claim 1, also characterized in that the juxtaposition of said low and high frequency bands covers the whole bandwidth of the voice signal to be compressed.

3. Process for transcoding a sampled and digitalized voice signal, characterized in that it includes the following steps:

splitting the voice signal into blocks of samples;

splitting the frequency band of said voice signal into a low frequency band and a high frequency band, coding the information of said low frequency band by splitting the band thereof into several sub-bands and by requantizing each of said sub-bands with a bit rate determined for each block of samples in accordance with the energy contained in said sub-band, said coding of the whole low frequency band supplying SIGNAL information, processing the contents of said high frequency band to generate a set of preemphasized parcor type coefficients (COEF) for each block of samples being processed, determining an energy information (ENERG) about the high frequency portion of the bandwidth of the signal to be compressed considered by blocks, and multiplexing the said SIGNAL, COEF and ENERG data on a transmission line.

4. A device for digitally compressing a voice signal, characterized in that it includes:

an input, first means for applying to said input the samples of a telephone band of the voice signal taken by blocks of constant length, second means connected to said input for receiving and splitting the applied signal bandwidth into a high frequency band and a low frequency band, third means responsive to the information of said high frequency band and generating the preemphasized parcor type coefficients (COEF) of the signal it contains, periodic resampling means receiving the information of said low band, submitting it to a periodic resampling operation and quantizing it by splitting it into sub-bands and dynamically adapting the bit rate assigned to the quantization of each sub-band in accordance with the energy contained in said sub-band, so as to generate a SIGNAL information, means for determining the energy of the high frequency band connected to said input and to said periodic resampling means and generating information ENERG, means receiving said information COEF, SIGNAL and ENERG for multiplexing them on a transmission line.

5. A device for the digitally compressing a voice signal in accordance with claim 4 characterized in that the determination of said preemphasized parcor type coefficients is carried out by using means including:

an input terminal, first means for applying the samples of the signal to be processed to said input terminal, means generating autocorrelation functions to determine a set of correlation coefficients for each of said blocks of samples, said means being connected to said input terminal, means for generating the ratio between the first two terms of said set of coefficients, a pre-emphasis filter based on said ratio, the input of which is connected to the output of said means for generating autocorrelation functions, partial autocorrelation means the input of which is connected to the output of said pre-emphasis filter, and the output of which delivers said partial correlation coefficients.

11. A data compression process for compressing a voice signal extending over the telephone bandwidth, which has been sampled at a first given rate and quantized with a relatively high number of bits, said process including the following steps:

splitting the frequency bandwidth of said sampled and quantized voice signal into a low frequency bandwidth and a high frequency bandwidth;

periodically resampling said low frequency bandwidth at a rate lower than said first given rate;

splitting the resampled low frequency bandwidth into p sub-bands;

subjecting each of said p sub-bands to an adaptive sub-band block requantizing operation using a quantizing step dynamically adapted to the sub-band energy contained in said each sub-band;

collecting the requantized information over the p sub-bands into a SIGNAL data information;

measuring and quantizing the energy contained in said high frequency bandwidth for deriving ENERG data;

processing the information contained in said telephone bandwidth for deriving COEF data representing preemphasized parcor type coefficients; and

,

multiplexing said SIGNAL, COEF and ENERG data.

6. Process for decoding a compressed coded voice signal of the character described in claims 11 or 12 characterized in that it includes the following steps:

separating through a demultiplexing operation, the data of channels ENERG, SIGNAL and COEF from each other and decoding said information;

performing a linear interpolation of the decoded information of the SIGNAL channel;

splitting said information into two paths, one direct and one indirect;

processing said indirect path by performing distortion, high pass filtering and normalization of the energy level contained in said indirect path by the ENERG information;

performing an operation of filtering the information delivered by the output of said indirect path through an inverse filter the coefficients of which are obtained from information COEF;

adding the contents of said direct path to the result of said filtering operation.

7. Decoding process according to claim 6 characterized in that a white noise energy is added to the signal appearing on said indirect path after said distortion.

8. Decoding process according to claim 7 also characterized in that the level of said white noise energy is dynamically adjusted in accordance with the energy level measured on said indirect path.

9. A device for decoding a compressed coded voice signal of the character described in claim 6 including:

input means;

demultiplexing means connected to said input means for separating the ENERG, SIGNAL and COEF data into a first, second and third channel respectively;

first, second and third decoding means, the inputs of which are connected to said first, second and third channels, respectively;

interpolation means the input of which is connected to the output of the decoding means of said second channel and the output of which feeds two paths;

one direct and one indirect;

means connected to said indirect path and including;

non-linear distortion means responsive to the interpolation means output;

white noise generating means for providing a white noise-like signal;

first adding means the inputs of which are connected to the outputs of said non-linear distortion means and said white noise generating means;

filtering and decimation means the input of which is connected to the output of said first adding means;

modulation means the inputs of which are connected to the output of said filtering and decimation means and to the output of said first decoding means;

inverse filtering means the input of which is connected to the output of said third decoding means and the coefficients of which are provided by the output of said modulation means; and

, second adding means for reconstructing the voice signal by adding the output of said inverse filtering means to the output of said interpolating means on said direct path.

10. A device for decoding a compressed coded voice signal according to claim 9 also characterized in that said modulation means include:

measuring means for measuring energy E2 provided by said inverse filtering means, comparing means for comparing E2 to the energy E1 provided by said first decoding means, and modulation means for modulating the energy provided by said white noise generating means by a coefficient selected from the information provided by said comparing means.

13. A data compression process according to claims 11 or 12 wherein said adaptive sub-band block requantizing operation uses the Block Companded PCM techniques performed over blocks of signal samples provided on the p sub-bands over a given period of time.

14. A data compression process according to claim 13, wherein the SIGNAL data includes p values of quantizing steps, p values of scale factors and the requantized signal samples on the p sub-bands over said given period of time.

12. A data compression process for compressing a voice signal extending over the telephone bandwidth, sampled at a first given rate and quantized with a relatively high number of bits, said process including the following steps:

splitting the frequency bandwith of said sampled and quantized voice signal into a low frequency bandwidth and a high frequency bandwidth;

periodically resampling said low frequency bandwith at a rate lower than said first given rate;

splitting the resampled low frequency bandwidth into p sub-bands;

subjecting each of said p sub-bands to an adaptive sub-band block requantizing operation using a quantizing step dynamically adapted to the sub-band energy contained in said each sub-band;

collecting the requantized information over the p sub-bands into a SIGNAL data information;

measuring and quantizing the energy contained in said high frequency bandwidth for deriving ENERG data;

processing the information contained in said high frequency bandwidth for deriving COEF data representing preemphasized parcor type coefficients; and

, multiplexing said SIGNAL, COEF and ENERG data.

15. A device for digitally compressing a voice signal extending over the telephone frequency bandwidth, said signal being sampled at a first given rate and quantized with a relatively high number of bits, said device including:

filtering means for receiving said sampled and quantized voice signal and providing the high frequency components of said voice signal at an output thereof and the low frequency components of said voice signal at another output thereof;

means for determining preemphasized parcor type coefficients receiving said low frequency components of said voice signal and providing preemphasized parcor type coefficients;

first requantizing means for receiving said preemphasized parcor type coefficients and providing COEF data;

periodic resampling means for receiving said high frequency components of said voice signal and for periodically resampling said high frequency components of said voice signal;

adaptive sub-band block requentizing means for receiving the periodically resampled high frequency components of said voice signal and providing SIGNAL data;

high band energy measuring means for receiving said high frequency components of said voice signal and providing information relative to the energy of said high frequency components of said voice signal;

second requantizing means for receiving said information relative to the energy of said high frequency components of said voice signal and providing ENERG data; and

multiplexing means for multiplexing said COEF, SIGNAL and ENERG data over a single output channel.

16. A device for digitally compressing a voice signal in accordance with claim 15 characterized in that the determination of said preemphasized parcor type coefficients is carried out by using means including:

an input terminal;

first means for applying the samples of the signal to be processed to said input terminal;

autocorrelation function generator means with input connected to said input terminal;

preemphasis filter means with input connected to the output of said autocorrelation function generator means; and

,

partial autocorrelation circuit means with input connected to the output of said preemphasis filter means.