Process and device for blind equalization of the effects of a transmission channel on a digital speech signal

• US 6,157,909 A
• Filed: 07/20/1998
• Issued: 12/05/2000
• Est. Priority Date: 07/22/1997
• Status: Expired due to Fees

• Current Assignee: Orange S.A.
• Original Assignee: Orange S.A.
• Inventors: Mauuary, Laurent, Monne, Jean
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Storm, Donald L.
• Application Number:
  US09/119,279
• Time in Patent Office: 869 Days
• Field of Search: 704/226-228, 704/204, 704/234, 375/232
• US Class Current: 704/228
• CPC Class Codes: G10L 21/0208 Noise filtering
• CPC Class Codes: G10L 21/0232 Processing in the frequency domain
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum
• CPC Class Codes: H04L 25/0307 using blind adaptation

Abstract

A process and device for blind equalization of the effects of a transmission channel on a speech signal. The speech signal is transformed into cepstral vectors which are representative of the speech signal over a given horizon. A reference cepstrum consisting of a constant cepstrum signal representative of the long-term cepstrum of the speech signal is calculated for each cepstral vector. Each of the cepstral vectors is subjected to adaptive filtering by LMS on the basis of the reference cepstrum so as to generate a set of equalized cepstral vectors on the basis of the calculation of an error signal between the reference cepstrum and equalized cepstral vectors. The error signal is expressed as the difference between the reference cepstrum component of a given rank and the component of the same rank of the equalized cepstral vector.

Claims

1. Process for the blind equalisation of the effects of a transmission channel on a digital speech signal ({s_n (t)}) passing through said transmission channel, said process involving at least:

transforming said digital speech signal into a set of cepstral vectors, the set of cepstral vectors being representative of said digital speech signal over a given horizon;

calculating a reference cepstrum ({R_n (i)}) representative, for each of the cepstral vectors of said set of cepstral vectors of the long-term cepstrum of said speech signal, said reference cepstrum ({R_n (i)}) consisting of a constant cepstrum signal for each of the cepstral vectors representative in the digital speech signal and in the set of cepstral vectors of the long-term cepstrum in the speech signal;

subjecting each of the cepstral vectors to adaptive filtering on the basis of said reference cepstrum, in order to generate a set of equalised cepstral vectors ({C_n (i)}) in which the effects of said transmission channel are substantially suppressed, said set of equalised cepstral vectors being representative of an equalised digital speech signal.

2. Process according to claim 1, wherein the cepstral transformation involves successively:

subjecting said digital speech signal ({s_n (t)}) to a frequency transform transmitting a frequency spectrum ({S_n (f)}) of said digital speech signal ({s_n (t)}) over the horizon under consideration;

decomposing said frequency spectrum ({S_n (f)}) into frequency sub-bands to generate a plurality of signals in frequency sub-bands ({V_n)(j)});

subjecting each signal in frequency sub-bands to logarithmic attenuation to generate a plurality of attenuated signals in frequency sub-bands ({LV_n (j)});

subjecting the attenuated set of signals in frequency sub-bands to reverse frequency transformation to generate said set of cepstral vectors ({C_n (i)}).

3. Process according to claim 1, wherein the reference cepstrum ({R_n (i)}) has a constant value which is identical for each component or cepstral coefficient (C_n (i)) of corresponding rank i, constituting each cepstral vector ({C_n (i)}).

4. Process according to claim 1, wherein said adaptive filtering involves, for each of the cepstral vectors on the basis of said reference cepstrum ({R_n (i)}) representative of the long-term cepstrum of this speech signal, for each of the cepstral vectors, carrying out least mean square (LMS) type filtering, said LMS type filtering enabling said equalised cepstral vectors to be generated on the basis of the calculation of error between said reference cepstrum ({R_n (i)}) and the set of said equalised cepstral vectors.

5. Device for the blind equalisation of the effects of a transmission channel on a digital speech signal ({s_n (t)}) passing through said transmission channel, comprising at least:

means for transforming said digital speech signal into a set of cepstral vectors, said set of cepstral vectors being representative of said digital speech signal over a given horizon;

means for generating a reference cepstrum that is representative, for each of the cepstral vectors of said set of cepstral vectors, of the long-term cepstrum of this speech signal;

means for adaptive filtering, on the basis of said reference cepstrum, of each of the cepstral vectors enabling generation of a set of equalised cepstral vectors in which the effect of said transmission channel are substantially suppressed, said set of equalised cepstral vectors being representative of an equalised digital speech signal, said means for adaptive filtering comprising at least for each cepstral vector ({C_n (i)}), iε and

for a reference cepstrum ({R_n (i)}) representative, for each of the cepstral vectors, of the long-term cepstrum of this speech signal,

means for calculating an error signal (E(i)) between each component of rank i (C_n (i)) of each equalised cepstral vector ({C_n (i)}) and the corresponding component of identical rank (R_n (i)) of the reference cepstrum ({R_n (i)}), E(i)=R _n (i)-C_n (i)); and

means for equalisation of said cepstral vector ({C_n (i)}) transmitting a component (C_n (i)) of the equalised cepstral vector ({C_n (i)}) on the basis of each component (C_n (i)) of each cepstral vector ({C_n (i)}) and said error signal (E(i)).

6. Device according to claim 5, wherein said means of transformation of said digital speech signal into a set of cepstral vectors comprise at least:

means for the frequency transform of said digital speech signal transmitting a frequency spectrum of said digital speech signal ({s_n (t)}) over the horizon under consideration;

a bank of filters for decomposing, into N frequency sub-bands, said frequency spectrum of said digital speech signal transmitting N signals in frequency sub-bands ({V_n (j)});

means of logarithmic attenuation of each signal in frequency sub-bands transmitting a plurality of attenuated signals in frequency sub-bands ({LV_n (j)});

means for reverse frequency transform enabling said set of cepstral vectors ({C_n (i)}) to be generated on the basis of said attenuated signals in frequency sub-bands ({LV_n (j)}).

7. Device according to claim 5, wherein said means for calculating the error signal (E(i)) and said means for equalisation of said cepstral vector ({C_n (i)}) comprise, for each component (C_n (i)) of said cepstral vector, a subtractor circuit receiving said component (R_n (i)) of the reference cepstrum ({R_n (i)}) and said component (C_n (i)) of said equalised cepstral vector ({C_n (i)}) and transmitting said error signal (E(i));

means for adaptation of this error signal (E(i)) comprising;

a circuit which multiplies by a multiplying coefficient μ

transmitting a weighted error signal (E*(i));

an adder circuit with two inputs and one output, a first input receiving said weighted error signal (E*(i)) and a second input receiving the signal transmitted by said first adder circuit via a slow-rise circuit that delays by a predetermined duration, the output of said slow-rise circuit transmitting an adaptation signal (H_n (i)); and

an equalising adder circuit receiving said cepstral coefficient (C_n (i)) and said adaptation signal (H_n (i)) and transmitting said equalised cepstral coefficient (C_n (i)).