Method for processing speech signal data and finding a filter coefficient
View Patent ↗Method and computing apparatus for processing speech signal data. A speech signal is divided into frames. Each frame is characterized by a frame number T representing a unique interval of time. Each speech signal is characterized by a power spectrum with respect to frame T and frequency band ω. A speech segment and a reverberation segment of the speech signal is determined. L filter coefficients W(k) (k=1, 2, . . . , L) respectively corresponding to L frames immediately preceding frame T are computed such that the L filter coefficients minimize a function Φ that is a linear combination of sum of squares of a residual speech power in the reverberation segment and a sum of squares of a subtracted speech power in the speech segment. The computed L filter coefficients are stored within storage media of the computing apparatus.
1. A computer-implemented method for determining a first filter coefficient in a reverberation rejection technique comprising:
obtaining the first filter coefficient in a reverberation rejection technique in which a second filter coefficient multiplied by a speech power spectrum in a past frame is subtracted from a speech spectrum in a current frame;
utilizing the first filter coefficient with a computer processor so as to minimize a weighted summation with the speech power spectrum, wherein the first filter coefficient is the second filter coefficient number of the speech power spectrum in the past frame, subtracted from the speech power spectrum of a current frame; and
determining the speech power spectrum in the current frame in a speech end reverberation segment where a fluctuation of the speech power spectrum is gradual compared to the case of no reverberation in the speech interval, and displaying the speech power spectrum on a computer display apparatus.
2. The method according to claim 1 , wherein the speech end reverberation segment is obtained by
obtaining a predetermined speech power track, whose speed following a speech power changes according to a level of the speech power, and by
determining that a difference between the predetermined speech power track and a power track of the current frame, which has been smoothed in a temporal direction, becomes greater than a predetermined threshold value.
3. The method according claim 1 , wherein the weighted summation is a weighted summation with a square of a subtracted speech power in a speech segment and a square of a residual speech power in the speech end reverberation segment.
4. The method according to claim 2 , wherein the predetermined speech power track is obtained with S(T) in the following Expression 1, and the speech frame in the current frame, which has been smoothed in the temporal direction, is obtained P(T) in the following Expression 1:
energy
(
T
)
=
10.0
*
log
1
0
(
1
N
∑
i
=
1
N
x
[
i
]
2
)
P
(
T
)
=
10
C
1
*
energy
(
T
)
Q
(
T
)
=
(
1
-
α
l
)
*
Q
(
T
-
1
)
+
α
l
*
P
(
T
)
α
l
=
C
2
*
C
3
*
Q
(
T
-
1
)
2
P
(
T
)
2
S
(
T
)
=
(
1
-
α
h
)
*
S
(
T
-
1
)
+
α
h
*
P
(
T
)
α
h
=
C
3
*
P
(
T
)
2
Q
(
T
-
1
)
2
wherein, X[i] is a temporal region speech data in a frame number; N is a total number of samples of the temporal region speech data in the frame number T; and C 1 , C 2 and C 3 are arbitrary constants.
5. The method according to claim 1 , wherein the first filter coefficient is determined by storing the second filter coefficient due to speech prior to a current speech.
6. The method according to claim 1 , wherein the first filter coefficient is determined by storing a current speech and by using a speech after the completion of the speech.
7. The method according to claim 1 , wherein the first filter coefficient is determined by sequentially updating the second filter coefficient every time a power spectrum of newly-observed speech is obtained.