Decision-directed frame-synchronous adaptive equalization filtering of a speech signal by implementing a hidden markov model
First Claim
1. A method of equalizing a speech signal constituted by an observed sequence of successive input sound frames, which speech signal is liable to be affected by disturbances, said method comprising a preliminary step of modelizing the speech signal by a hidden Markov model and, at each instant t, the further steps of:
- (a) forming a plurality of equalization filters associated with paths in the Markov sense at instant t, on the basis of a plurality of earlier sound frames and on the basis of a plurality of parameters of the Markov model;
(b) applying at least a plurality of said equalization filters to said frames to obtain, at instant t, a plurality of filtered sound frames sequences and an utterance probability for each of the paths respectively associated with the equalization filters applied;
(c) selecting that one of said equalization filters which corresponds to a most probable path, in the Markov sense, which is that of said paths which has a highest utterance probability of the sequence of sound frames filtered by the corresponding applied equalization filter; and
(d) selecting as an equalized frame the filtered frame supplied by the selected equalization filter.
1 Assignment
0 Petitions
Accused Products
Abstract
For equalizing a speech signal constituted by an observed sequence of successive input sound frames, which speech signal is liable to be affected by disturbances, the speech signal is modelled by means of a hidden Markov model and, at each instant t: equalization filters are constituted in association with the paths in the Markov sense at instant t; at least a plurality of the equalization filters are applied to the frames to obtain, at instant t, a plurality of filtered sound frame sequences and an utterance probability for each of the paths respectively associated with the equalization filters applied; the equalization filter corresponding to the most probable path in the Markov sense is selected; and the filtered frame supplied by the selected equalization filter is selected as the equalized frame.
-
Citations
14 Claims
-
1. A method of equalizing a speech signal constituted by an observed sequence of successive input sound frames, which speech signal is liable to be affected by disturbances, said method comprising a preliminary step of modelizing the speech signal by a hidden Markov model and, at each instant t, the further steps of:
-
(a) forming a plurality of equalization filters associated with paths in the Markov sense at instant t, on the basis of a plurality of earlier sound frames and on the basis of a plurality of parameters of the Markov model; (b) applying at least a plurality of said equalization filters to said frames to obtain, at instant t, a plurality of filtered sound frames sequences and an utterance probability for each of the paths respectively associated with the equalization filters applied; (c) selecting that one of said equalization filters which corresponds to a most probable path, in the Markov sense, which is that of said paths which has a highest utterance probability of the sequence of sound frames filtered by the corresponding applied equalization filter; and (d) selecting as an equalized frame the filtered frame supplied by the selected equalization filter. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12)
-
5. A method according to claim 4, wherein the most probable path is selected as being that of said paths which satisfies the following approximation:
-
space="preserve" listing-type="equation">s(T.sub.f)=argmax.sub.s P(F.sub.θ
1(s) (x.sub.1), . . . , F.sub.θ
Tf(s) (x.sub.Tf),s/λ
)!,where θ
i (s) is a best parameter vector of the equalization filter whose function is F.sub.θ
i(s), estimated on the i first points of path s.
-
-
6. A method according to claim 3, wherein the most probable path is selected as being that of said paths which satisfies the following approximation:
-
space="preserve" listing-type="equation">s(t)=argmax.sub.s P(F.sub.θ
1(s) (x.sub.1), . . . , F.sub.θ
t(s) (x.sub.t),s/λ
)!,where θ
i (s) is a best parameter vector of the equalization filter whose function is F.sub.θ
i (s), estimated on the i first points of path s.
-
-
7. A method according to claim 2, wherein for each path, the equalization filter has the following function:
F.sub.θ
(x)=x-bwhere θ
=b, and b is given by ##EQU12##
-
8. A method according to claim 2, wherein said disturbances are filtered using linear multiple regression, by means of an equalization filter whose function is, for each path, an affine function:
F.sub.θ
(x)=A.x+b,where θ
=(A,b), where A is a square matrix and b is a column vector.
-
9. A method according to claim 8, wherein said matrix A is diagonal.
-
10. A method according to claim 1, wherein said input sound frames are cepstrum vectors which are computed from the speech signal on time windows which overlap partially.
-
11. A method according to claim 1, wherein said disturbances are of convolutive and/or of additive nature.
-
12. A method according to claim 1, wherein said disturbances vary little over time.
-
-
13. An apparatus for equalizing a speech signal constituted by an observed sequence of successive input sound frames, which speech signal is liable to be affected by disturbances, comprising
means for modelizing the speech signal by a hidden Markov model; -
means for forming a plurality of equalization filters associated with paths in the Markov sense at instant t, on the basis of a plurality of earlier sound frames and on the basis of a plurality of parameters of the Markov model; means for applying at least a plurality of said equalization filters to said frames to obtain, at instant t, a plurality of filtered sound frame sequences and an utterance probability for each of the paths respectively associated with the equalization filters applied; means for selecting that one of said equalization filters which corresponds to a most probable path, in the Markov sense, which is the path having a highest utterance probability of the sequence of sound frames filtered by the corresponding applied equalization filter; and means for selecting as an equalized frame the filtered frame supplied by the selected equalization filter, said apparatus receiving at its input the speech signal affected by disturbances and delivering at its output a speech signal with reduced disturbances. - View Dependent Claims (14)
-
Specification