Method and apparatus for recognizing speech in a noisy environment
AssigneeSRI International
InventorsVenkata Ramana Rao Gadde, Horacio Franco, John Butzberger
Life of this patent
Estimating the term…
Abstract
An apparatus and a concomitant method for speech recognition. In one embodiment, the present method is referred to as a “Dynamic Noise Compensation” (DNC) method where the method estimates the models for noisy speech using models for clean speech and a noise model. Specifically, the model for the noisy speech is estimated by interpolation between the clean speech model and the noise model. This approach reduces computational cycles and does not require large memory capacity.
Claims12
Method for performing speech recognition on an input audio signal having a speech component and a noise component, said method comprising the steps of:
- (a) obtaining at least one clean speech model;
- (b) obtaining at least one noise model;
- (c) estimating a signal-to-noise ratio of the input audio signal;
- (d) generating a weight in accordance with the signal-to-noise ratio by accessing a signal-to-noise ratio/weight table;
- (e) applying said weight to said at least one noise model and said at least one clean speech model to derive said at least one noisy speech model;
- and
- (f) applying said at least one noisy speech model to extract a recognized text from the input audio signal.
Apparatus for performing speech recognition on an input audio signal having a speech component and a noise component, said apparatus comprising:
- means for obtaining at least one clean speech model;
- means for obtaining at least one noise model;
- means for estimating a signal-to-noise ratio of the input audio signal;
- means for generating a weight in accordance with said signal-to-noise ratio by accessing a signal-to-noise ratio/weight table;
- means for applying said weight to said at least one noise model and said at least one clean speech model to derive said at least one noisy speech model;
- and
- means for applying said at least one noisy speech model to extract a recognized text from the input audio signal.
A computer-readable medium having stored thereon a plurality of instructions, the plurality of instructions including instructions which, when executed by a processor, cause the processor to perform the steps of a method for performing speech recognition on an input audio signal having a speech component and a noise component, said method comprising the steps of:
- (a) obtaining at least one clean speech model;
- (b) obtaining at least one noise model;
- (c) estimating a signal-to-noise ratio of the input audio signal;
- (d) generating a weight in accordance with the signal-to-noise ratio by accessing a signal-to-noise ratio/weight table;
- (e) applying said weight to said at least one noise model and said at least one clean speech model to derive said at least one noisy speech model;
- and
- (f) applying said at least one noisy speech model to extract a recognized text from the input audio signal.
Dependencies are inferred from the claim text.