Apparatus and method for noise attenuation in a speech recognition system
First Claim
Patent Images
1. An apparatus for noise attenuation in an electronic system, comprising:
- a noise suppressor configured to selectively attenuate additive noise in an electronic signal, said electronic signal being a noisy speech signal that includes a noise signal combined with a speech signal, said noise suppressor selectively attenuating said noise signal by utilizing statistical characteristics of amplitude energy values of said noise signal, said statistical characteristics of said amplitude energy values of said noise signal include a noise average and a noise standard deviation, said noise suppressor generating an attenuated noisy speech signal according to a formula;
where Yatk is said attenuated noisy speech signal for a frequency k, Yk is said noisy speech signal for said frequency k, μ
k is said noise average for said frequency k, σ
k is said noise standard deviation for said frequency k, α
is a overestimation coefficient, and A is an attenuation coefficient; and
a processor coupled to said electronic system to control said noise suppressor.
1 Assignment
0 Petitions
Accused Products
Abstract
The noise suppressor utilizes statistical characteristics of the noise signal to attenuate amplitude values of the noisy speech signal that have a probability of containing noise. In one embodiment, the noise suppressor utilizes an attenuation function having a shape determined in part by a noise average and a noise standard deviation. In a further embodiment, the noise suppressor also utilizes an adaptive attenuation coefficient that depends on signal-to-noise conditions in the speech recognition system.
250 Citations
39 Claims
-
1. An apparatus for noise attenuation in an electronic system, comprising:
-
a noise suppressor configured to selectively attenuate additive noise in an electronic signal, said electronic signal being a noisy speech signal that includes a noise signal combined with a speech signal, said noise suppressor selectively attenuating said noise signal by utilizing statistical characteristics of amplitude energy values of said noise signal, said statistical characteristics of said amplitude energy values of said noise signal include a noise average and a noise standard deviation, said noise suppressor generating an attenuated noisy speech signal according to a formula;
where Yatk is said attenuated noisy speech signal for a frequency k, Yk is said noisy speech signal for said frequency k, μ
k is said noise average for said frequency k, σ
k is said noise standard deviation for said frequency k, α
is a overestimation coefficient, and A is an attenuation coefficient; and
a processor coupled to said electronic system to control said noise suppressor. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10)
where Ak(t) is said adaptive attenuation coefficient for a frequency index k at a frame t, A is said attenuation coefficient, α
is said overestimation coefficient, μ
k(t) is said noise average for frequency index k at frame t, Spk(t) is a noisy speech average for frequency index k at frame t, μ
k(prev) is a noise average for a noise period immediately previous to a current utterance, and Spk(prev) is a noisy speech average for an utterance immediately previous to a current noise period.
-
-
7. The apparatus of claim 6, wherein said noise suppressor calculates said noisy speech average according to a formula:
-
8. The apparatus of claim 1, wherein said noise suppressor determines a noise average and a noise standard deviation of said energy amplitude values of said noise signal, utilizes said noise average and said noise standard deviation to identify selected ones of said amplitude energy values of said noisy speech signal that have a probability of containing noise, and selectively attenuates said amplitude energy values of said noisy speech signal according to said probability.
-
9. The apparatus of claim 1, wherein said noise suppressor calculates said noise average according to a formula:
-
where μ
k is said noise average for a frequency index k, Nk(t) is a noise energy amplitude value for frequency index k at a frame t for t equal to 1 through T, and T is a total number of frames in a noise period.
-
-
10. The apparatus of claim 9, wherein said noise suppressor calculates said noise standard deviation according to a formula:
-
where σ
k is said noise standard deviation for frequency index k, μ
k is said noise average for frequency index k, Nk(t) is said noise energy amplitude value for frequency index k at said frame t for t equal to 1 through T, and T is said total number of frames in said noise period.
-
-
11. An apparatus for noise attenuation in an electronic system, comprising:
-
a noise suppressor configured to selectively attenuate additive noise in an electronic signal, said electronic signal being a noisy speech signal that includes a noise signal combined with a speech signal, said noise suppressor selective attenuating said noise signal by utilizing statistical characteristics of amplitude energy values of said noise signal, said statistical characteristics of said amplitude energy values of said noise signal include a noise average and a noise standard deviation, said noise suppressor generating an attenuated noisy speech signal according to a formula;
where Yatk is said attenuated noisy speech signal for a frequency k, Yk is said noisy speech signal for said frequency k, μ
k is said noise average for said frequency k, σ
k is said noise standard deviation for said frequency k, α
v is an overestimation coefficient related to said noise standard deviation, and A is an attenuation coefficient; and
a processor coupled to said electronic system to control said noise suppressor.
-
-
12. An apparatus for noise attenuation in an electronic system, comprising:
a noise suppressor configured to selectively attenuate additive noise in an electronic signal, said electronic signal being a noisy speech signal that includes a noise signal combined with a speech signal, said noise suppressor selectively attenuating said noise signal by utilizing statistical characteristics of amplitude energy values of said noise signal, said statistical characteristics of said amplitude energy values of said noise signal include a noise average and a noise standard deviation, said noise suppressor calculating said noise average according to a formula;
- View Dependent Claims (13, 14)
-
15. A method for noise attenuation in an electronic system, comprising the steps of:
-
selectively attenuating additive noise in an electronic signal using a noise suppressor, said electronic signal being a noisy speech signal that includes a noise signal combined with a speech signal, said noise suppressor selectively attenuating said noise signal by utilizing statistical characteristics of amplitude energy values of said noise signal, said statistical characteristics of said amplitude energy values of said noise signal include a noise average and a noise standard deviation, said noise suppressor generating an attenuated noisy speech signal according to a formula;
where Yatk said attenuated noisy speech for signal for a frequency k, Yk is said noisy speech signal for said frequency k, μ
k is said noise average for said frequency k, σ
k is said noise standard deviation for said frequency k, α
is a overestimation coefficient, and A is an attenuation coefficient; and
controlling said noise suppressor with a processor coupled to said electronic system. - View Dependent Claims (16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32)
where Ak(t) is said adaptive attenuation coefficient for a frequency index k at a frame t, A is said attenuation coefficient, α
is said overestimation coefficient, μ
k(t) is said noise average for frequency index k at frame t, Spk(t) is a noisy speech average for frequency index k at frame t, μ
k(prev) is a noise average for a noise period immediately previous to a current utterance, and Spk(prev) is a noisy speech average for an utterance immediately previous to a current noise period.
-
-
21. The method of claim 20, wherein said noise suppressor calculates said noisy speech average according to a formula:
-
22. The method of claim 15, wherein said noise suppressor determines a noise average and a noise standard deviation of said energy amplitude values of said noise signal, utilizes said noise average and said noise standard deviation to identify selected ones of said amplitude energy values of said noisy speech signal that have a probability of containing noise, and selectively attenuates said amplitude energy values of said noisy speech signal according to said probability.
-
23. The method of claim 15, wherein said noise suppressor calculates said noise average according to a formula:
-
where μ
k is said noise average for a frequency index k, Nk(t) is a noise energy amplitude value for frequency index k at a frame t for t equal to 1 through T, and T is a total number of frames in a noise period.
-
-
24. The method of claim 23, wherein said noise suppressor calculates said noise standard deviation according to a formula:
-
where σ
k is said noise standard deviation for frequency index k, μ
k is said noise average for frequency index k, Nk(t) is said noise energy amplitude value for frequency index k at said frame t for t equal to 1 through T, and T is said total number of frames in said noise period.
-
-
25. The method of claim 15, further comprising the step of generating amplitude energy values of said noisy speech signal using a Fast Fourier transformer.
-
26. The method of claim 25, further comprising the steps of providing attenuated noisy speech amplitude energy values to a filter bank, and generating channel energies using said filter bank.
-
27. The method of claim 26, further comprising the step of converting said channel energies into logarithmic channel energies using a logarithmic compressor.
-
28. The method of claim 27, further comprising the step of converting said logarithmic channel energies into corresponding static features using a frequency cosine transformer.
-
29. The method of claim 28, further comprising the step of providing said corresponding static features to a normalizer, a first cosine transformer, and a second cosine transformer.
-
30. The method of claim 29, further comprising the steps of converting said corresponding static features into delta features using said first cosine transformer, converting said corresponding static features into delta-delta features using said second cosine transformer, and providing said delta features and said delta-delta features to said normalizer.
-
31. The method of claim 30, further comprising the step of normalizing said static features, said delta features, and said delta-delta features using said normalizer to produce normalized static features, normalized delta features, and normalized delta-delta features.
-
32. The method of claim 31, further comprising the step of analyzing said normalized static features, said normalized delta features, and said normalized delta-delta features using a recognizer to produce a speech recognition result.
-
33. A method for noise attenuation in an electronic system, comprising the steps of:
-
selectively attenuating additive noise in an electronic signal using a noise suppressor, said electronic signal being a noisy speech signal that includes a noise signal combined with a speech signal, said noise suppressor selectively attenuating said noise signal by utilizing statistical characteristics of amplitude energy values of said noise signal, said statistical characteristics of said amplitude energy values of said noise signal include a noise average and a noise standard deviation, said noise suppressor generating an attenuated noisy speech signal according to a formula;
where Yatk is said attenuated noisy speech signal for a frequency k, Yk is said noisy speech signal for said frequency k, μ
k is said noise average for said frequency k, σ
k is said noise standard deviation for said frequency k, α
v is an overestimation coefficient related to said noise standard deviation, and A is an attenuation coefficient; and
controlling said noise suppressor with a processor coupled to said electronic system.
-
-
34. A method for noise attenuation in an electronic system, comprising the steps of:
selectively attenuating additive noise in an electronic signal using a noise suppressor, said electronic signal being a noisy speech signal that includes a noise signal combined with a speech signal, said noise suppressor selectively attenuating said noise signal by utilizing statistical characteristics of amplitude energy values of said noise signal, said statistical characteristics of said amplitude energy values of said noise signal include a noise average and a noise standard deviation, said noise suppressor calculating said noise average according to a formula;
- View Dependent Claims (35, 36)
-
37. An apparatus for noise attenuation in an electronic system, comprising:
a noise suppressor configured to selectively attenuate additive noise in an electronic signal, said noise suppressor determining a noise average and a noise standard deviation of energy amplitude values of a noisy speech signal, said noise suppressor utilizing said noise average and said noise standard deviation to identify said amplitude energy values of said noisy speech signal that have a statistical probability of containing said additive noise, said noise suppressor selectively attenuating said amplitude energy values of said noisy speech signal according to said statistical probability, said noise suppressor calculating said noise average according to a formula;
- View Dependent Claims (38, 39)
Specification