Process and device for creating comfort noise in a digital speech transmission system

• US 5,812,965 A
• Filed: 10/11/1996
• Issued: 09/22/1998
• Est. Priority Date: 10/13/1995
• Status: Expired due to Term

• Current Assignee: Orange S.A.
• Original Assignee: Orange S.A.
• Inventors: Massaloux, Dominique
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): EDOUARD, PATRICK NESTOR
• Application Number:
  US08/731,381
• Time in Patent Office: 711 Days
• Field of Search: 704/205, 704/226, 704/227, 704/228, 704/219, 704/201, 704/214, 704/220
• US Class Current: 704/205
• CPC Class Codes: G10L 19/012 Comfort noise or silence coding
• CPC Class Codes: G10L 25/18 the extracted parameters being spectral information of each sub-band
• CPC Class Codes: H04J 3/175 Speech activity or inactivity detectors echo suppressors H04B3/20

Abstract

A process and a device for creating comfort noise for discontinuous speech transmission in which, following the end of each detected active period, a silence descriptor frame is generated as a set of parameters describing the comfort noise. For every successive current inactive frame, the frequency spectrum is analysed and then compared with a reference spectrum stored previously. Upon identity of the reference frequency spectrum and the current frame spectrum, no new silence descriptor frame is transmitted. Upon absence of identity of these spectra, a new silence descriptor frame is transmitted.

Claims

1. A process for creating comfort noise in a system for the digital transmission of discontinuous speech, in which the coded speech signal is transmitted during an active period, during which active frames are transmitted, each active period being followed by an inactive period, during which at least inactive frames are transmitted, wherein said process consists, at the sending end, on detecting an inactive period, in:

generating and transmitting a silence descriptor frame, consisting of a set of coded parameters describing the comfort noise, said silence descriptor frame constituting a first silence descriptor frame of said inactive period following said active period; and

for every successive current inactive frame of said inactive period;

analysing and storing the frequency spectrum of said current inactive frame;

comparing the frequency spectrum of said current inactive frame with a reference frequency spectrum, and upon criterion of identity of the reference and current inactive frame frequency spectra;

deferring any transmission, especially that of a new silence descriptor frame during said current inactive frame, and upon criterion of absence of identity of said reference and current inactive frame frequency spectra;

generating and transmitting a new silence descriptor frame during said current inactive frame, thereby allowing to reduce the transmission rate of the comfort noise to that of merely the silence descriptor frames whose frequency spectrum is different from said reference spectrum estimated during the preceding silence descriptor frame.

2. The process of claim 1, wherein the reference frequency spectrum is formed from the frequency spectrum of the ambient noise estimated and stored during the creation of each silence descriptor frame.

3. The process for creating comfort noise in a system for the digital transmission of discontinuous speech, in which the speech signal is transmitted during an active period, during which active frames are transmitted, each active period being followed by an inactive period, during which at least inactive frames are transmitted, silence descriptor frames being transmitted at the sending end according to claim 1, wherein said process consists, at the receiving end, in conjunction with a speech decoder, in:

decoding, for each successive silence descriptor frame, a set of coded parameters describing the comfort noise, in order to generate decoded parameters;

synthesizing, at the level of said speech decoder, after processing said decoded parameters, the corresponding comfort noise.

4. A process for synthesizing an excitation signal used in a comfort noise generator in a system for the digital transmission of discontinuous speech comprising a speech coder of predictive type, based on a specified number of excitation samples emanating from past frames and from long-term prediction data LTP delivered by this speech coder, wherein said process consists:

in subdividing each current inactive frame, including N samples, into N/L blocks each including N/L successive samples;

in randomly drawing the codes of the parameters of the long-term excitation, by using said excitation samples emanating from past frames, in order to obtain L long-term prediction excitation samples e_LTP (n);

in randomly drawing codes of an innovation waveform, in order to obtain L samples u(n) of said waveform;

in determining a gain value β

from the quantization index of said gain;

in determining, by updating the samples of each current block, an excitation signal Exc, said excitation signal being defined for said current frame as a linear combination, based on the gain value β

, of the long-term excitation e_LTP (n) and of said innovation waveform u(n).

5. The process of claim 4, wherein the value of the gain β

is associated with said innovation waveform, said gain value being obtained by selecting the quantization index allowing to obtain, over said current block, the per-sample mean energy which is as close as possible to a desired value g_t², upon a criterion of minimizing the expression;

##EQU10## said linear combination defining the excitation signal Exc satisfying the relation;

space="preserve" listing-type="equation">Exc(jL+k)=e.sub.LTP (jL+k)+β

u(jL+k)

with n=jL+k and k varying from 0 to L-1.

6. The process of claim 4, wherein said gain value β

is associated with said long-term excitation and with said innovation waveform, by weighting on the basis of respectively associated gain values, β

₁, β

₂, said gain values being obtained by selecting a quantization index allowing to obtain, over said current block, a per-sample mean energy value which is as close as possible to a given value g_t², upon a criterion of minimizing the expression;

##EQU11## said linear combination defining an excitation signal Exc satisfying the relation;

space="preserve" listing-type="equation">Exc(jL+k)=β

.sub.1 ·

e.sub.LTP (jL+k)+β

.sub.2 ·

u(jL+k)

with n=jL+k and k varying from 0 to L-1.

7. A device for creating comfort noise in a system for the digital transmission of discontinuous speech, in which the speech signal is transmitted during an active period, during which active frames are transmitted, each active period being followed by an inactive period, during which at least inactive frames are transmitted, said device including, at the sending end, constituting a comfort noise generator:

means for creating and conditionally transmitting, upon detecting an inactive period, a silence descriptor frame, consisting of a set of coded parameters describing the comfort noise, said silence descriptor frame constituting a first silence descriptor frame of said inactive period following said active period;

means for analysing and storing the frequency spectrum of every successive current inactive frame of said inactive period;

means for comparing the frequency spectrum of said current inactive frame with a reference frequency spectrum delivering a specified command signal upon criterion of identity and of absence of identity of said reference and current frame frequency spectra;

means for controlling the disabling and enabling respectively of the transmission of a new silence descriptor frame during said current inactive frame, receiving said specified command signal delivered by said comparing means, so as to defer any transmission especially that of a new silence descriptor frame upon the identity of said reference and current frame frequency spectra, and to carry out the transmission of a new silence descriptor frame upon the absence of identity of said reference and current frame frequency spectra, the rate of transmission of the comfort noise thus being reduced to that of merely the silence descriptor frames whose frequency spectrum is different from the reference spectrum estimated during a preceding silence descriptor frame.

8. The device of claim 7, wherein the digital transmission system including, at the sending end, a voice activity device delivering a voice activity signal, a predictive speech coder, said coder performing an analysis of the speech signal by means of an LPC filter delivering an autocorrelation function signal for the speech signal, a signal describing the LPC filtering parameters relating to the preceding frame, and, conditionally, an excitation signal of at least one preceding frame, said device further comprises, constituting the comfort noise generator:

a module for calculating parameters relating to the current frame receiving the autocorrelation function signal for the speech signal and producing signals corresponding to the parameters of a current frame;

a decision module for sending a frame of coded speech frame, a silence descriptor frame or for suspending sending, said decision module receiving said voice activity signal and said signals corresponding to the parameters of said current frame and to the reference parameters and delivering a frame transmission type signal having three levels, a first level corresponding to the total absence of sending during an inactive current frame, a second level corresponding to the sending of a coded speech frame during an active frame, and a third level corresponding to the sending of a silence descriptor frame during an inactive frame;

a module for the conditional calculation of reference parameters, said module receiving as input the autocorrelation function signal for the speech signal and said signals corresponding to the parameters of said current frame, and delivering a reference parameters signal;

a coded silence descriptor frame generator module receiving said transmission type signal, said signals corresponding to the reference parameters and conditionally delivering a coded silence descriptor frame signal;

a module generating an excitation signal receiving the transmission type signal, and, conditionally, the past excitation signal supplied by said speech coder and delivering conditionally on the frame transmission type signal an excitation signal for said current frame;

a module generating a signal describing the updated LPC filtering parameters, receiving said frame transmission type signal, a signal describing the LPC parameters of the preceding frames which is delivered by said speech coder, and delivering conditionally on said frame transmission type signal a signal describing the updated LPC parameters for said current frame.

9. A process for creating a silence descriptor frame in order to generate comfort noise in a system for the transmission of discontinuous speech in which the coded speech signal is transmitted during an active period, said transmission consisting thus of successive active periods during which coded speech signal frames are transmitted, interspersed with inactive periods during which no speech signal frame is transmitted, said inactive periods consisting thus of inactive frames, including at least one silence descriptor frame forming an initial one of said inactive frames and successive inactive frames preceding a current inactive frame, for which transmission of a new silence descriptor frame can take place, wherein, in order to create each silence descriptor frame, said process comprises the steps of:

measuring local stationarity of a frequency spectrum of a plurality of said successive inactive frames preceding said current inactive frame;

determining a mean frequency spectrum of said successive inactive frames in order to establish a past mean frequency spectrum and upon criterion of identity and of absence of identity of a past mean frequency spectrum and of the frequency spectrum of said current inactive frame; and

selecting, in order to compute said silence descriptor frame, said past mean frequency spectrum and said spectrum of said current inactive frame respectively, thereby allowing to estimate a silence spectrum for every silence descriptor frame by taking into account the fact that said current inactive frame belongs either to a locally stationary spectral zone or to a non-stationary spectral zone.

10. A device generating a comfort noise in a system for the digital transmission of discontinuous speech, in which a coded speech signal is transmitting during an active period, during which active coded speech frames are transmitted, each active period being followed by an inactive period, during which at least silence descriptor frames are transmitted at the sending end, with said silence descriptor frames being transmitted on detecting an inactive period by generating and transmitting a silence descriptor frame, consisting of a set of coded parameters describing the comfort noise, said silence descriptor frame constituting a first silence descriptor frame of said inactive period following said active period, and for every successive current inactive frame of said inactive period, analyzing and storing the frequency spectrum of said current inactive frame, comparing the frequency spectrum of said current inactive frame with a reference frequency spectrum, and upon criterion of identity of the reference and current inactive frame frequency spectra, deferring any transmission, including that of a new silence descriptor frame, during said current inactive frame, and upon criterion of absence of identity of said reference and current inactive frame frequency spectra, generating a transmitting a new silence descriptor frame during said current inactive frame, so as to reduce the transmission rate of the comfort noise to that of merely the silence descriptor frames whose frequency spectrum is different from said reference spectrum estimated during the preceding silence descriptor frame, wherein said device comprises, at a receiving end, means for decoding in each successive silence descriptor frame the set of coded parameters describing the comfort noise, said means receiving a cue relating to the type of current frame processed, a coded silence descriptor frame signal and, from a speech decoder, a management signal ensuring synchronous operation with said speech decoder, in order to generate decoded parameters including a signal describing the comfort noise relating to said current frame, said decoding means being interconnected with a speech signal decoder receiving, on the one hand, said coded speech frames, and, on the other hand, said signal describing the comfort noise relating to said current frame, said speech decoder delivering by synthesis a speech signal or a synthesized comfort noise.

11. The device of claim 10, wherein, in the case in which said speech signal decoder is formed by a predictive decoder, said decoding means include at least:

a demultiplexor receiving the transmitted coded silence descriptor frame signal and delivering a coded signal describing the LPC filtering parameters relating to the current frame and a quantized gain index signal describing the synthesized excitation signal;

a dequantizer module receiving the quantized gain index signal and delivering a dequantized gain signal;

a calculation module receiving the coded signal describing the LPC filtering parameters relating to said current frame, said signal describing the LPC filtering parameters relating at least to the preceding frame and the frame transmission type cue and delivering an updated signal with the LPC parameters relating to said current frame;

an excitation generator module receiving a dequantized gain signal, conditionally a past excitation signal and said frame transmission type cue and delivering an excitation signal relating to said current frame.