Frame erasure concealment technique for a bitstream-based feature extractor
View Patent ↗A frame erasure concealment technique for a bitstream-based feature extractor in a speech recognition system particularly suited for use in a wireless communication system operates to “delete” each frame in which an erasure is declared. The deletions thus reduce the length of the observation sequence, but have been found to provide for sufficient speech recognition based on both single word and “string” tests of the deletion technique.
1. A method, comprising:
measuring a Euclidean distance between line spectrum pairs of adjacent frames representative of speech signals, wherein the Euclidian distance is given by
∑
i
=
1
p
(
ω
n
,
i
-
ω
n
-
1
,
i
)
2
=
(
1
-
c
)
∑
i
=
1
p
(
ω
n
-
1
,
i
-
ω
d
c
,
i
)
2
,
wherein ω n,i represents an i-th line spectrum pair of an n-th frame, ω dc,i represents an empirical mean value of the i-th line spectrum pair, and c represents a forgetting factor; and
deleting a frame of the adjacent frames in response to determining that the Euclidean distance is one of less than and equal to a steady-state threshold given by T=(1−c)x 2 max [x 1 . . . x p ]εΩ Σ i=1 p (x i −ω dc,i ) 2 where Ω is a p-dimensional line spectrum pair vector space.
2. The method of claim 1 , further comprising:
computing a probability of an observation sequence based on a decoding algorithm in response to detecting a frame erasure of a particular speech signal.
3. The method of claim 2 , further comprising:
decoding the observation sequence based on a hidden Markov model process.
4. The method of claim 1 , wherein detecting the frame erasure comprises:
determining an error in bits most sensitive to error within a frame; and
declaring the frame erasure associated with the frame.
5. The method of claim 4 , wherein the bits most sensitive to error in a frame includes line spectrum pair information bits and gain information bits.
6. The method of claim 2 , wherein computing the probability of an observation sequence based on the decoding algorithm comprises:
determining a missing observation vector; and
computing the probability of the observation sequence using a missing feature theory based on the missing observation vector.
7. A method, comprising:
computing a probability of an observation sequence based on a decoding algorithm in response to detecting a frame erasure; and
deleting an observation vector of the observation sequence, the observation vector associated with the frame erasure, wherein the means for deleting the observation vector of the observation sequence comprises:
measuring a Euclidean distance between line spectrum pairs of contiguous frames, wherein the Euclidian distance is given by
∑
i
=
1
p
(
ω
n
,
i
-
ω
n
-
1
,
i
)
2
=
(
1
-
c
)
∑
i
=
1
p
(
ω
n
-
1
,
i
-
ω
d
c
,
i
)
2
,
wherein ω n,i represents an i-th line spectrum pair of an n-th frame, ω dc,i represents an empirical mean value of the i-th line spectrum pair, and c represents a forgetting factor; and
deleting a frame of the contiguous frames in response to determining that the Euclidean distance is one of less than and equal to a steady-state threshold given by T=(1−c)x 2 max [x 1 . . . x p ]εΩ Σ i=1 p (x i −ω dc,i ) 2 where Ω is a p-dimensional line spectrum pair vector space.
8. The method of claim 7 , further comprising:
decoding the observation sequence based on a Markov model process.
9. The method of claim 7 , wherein detecting the frame erasure comprises:
determining an error in bits most sensitive to error within a frame; and
declaring the frame erasure associated with the frame.
10. The method of claim 9 , wherein the bits most sensitive to error in a frame includes line spectrum pair information bits and gain information bits.
11. The method of claim 7 , wherein computing a probability of an observation sequence comprises:
determining a missing observation vector; and
computing the probability of the observation sequence using a missing feature theory based on the missing observation vector.
12. A method, comprising:
computing a probability of an observation sequence based on a decoding algorithm in response to detecting a frame erasure; and
measuring a Euclidean distance between line spectrum pairs of adjacent frames representative of speech signals, wherein the Euclidian distance is given by
∑
i
=
1
p
(
ω
n
,
i
-
ω
n
-
1
,
i
)
2
=
(
1
-
c
)
∑
i
=
1
p
(
ω
n
-
1
,
i
-
ω
d
c
,
i
)
2
,
wherein ω n,i represents an i-th line spectrum pair of an n-th frame, ω dc,i represents an empirical mean value of the i-th line spectrum pair, and c represents a forgetting factor; and
deleting a frame of the adjacent frames in response to determining that the Euclidean distance is less than to a steady-state threshold given by T=(1−c)x 2 max [x 1 . . . x p ]εΩ Σ i=1 p (x i −ω dc,i ) 2 where Ω is a p-dimensional line spectrum pair vector space.
13. The method of claim 12 , further comprising:
decoding the observation sequence based on a Markov model process.
14. The method of claim 12 , wherein detecting the frame erasure comprises:
determining an error in bits most sensitive to error within a frame; and
declaring the frame erasure associated with the frame.
15. The method of claim 14 , wherein the bits most sensitive to error in a frame includes line spectrum pair information bits and gain information bits.
16. The method of claim 12 , wherein computing a probability of an observation sequence comprises:
determining a missing observation vector; and
computing the probability of the observation sequence using a missing feature theory based on the missing observation vector.