Block-diagonal covariance joint subspace tying and model compensation for noise robust automatic speech recognition
First Claim
Patent Images
1. A noise robust automatic speech recognition system, comprising:
- a front end analysis module isolating a set of independent subspaces, wherein said front end analysis module employs one or more block diagonal front-end whitening matrices to isolate the set of independent subspaces;
a model-compensation module employing a model-compensation distortion function that operates on each of the subspaces isolated by said front-end analysis module; and
a subspace model compression module employing subspace tying to perform model compression.
7 Assignments
0 Petitions
Accused Products
Abstract
Model compression is combined with model compensation. Model compression is needed in embedded ASR to reduce the size and the computational complexity of compressed models. Model-compensation is used to adapt in real-time to changing noise environments. The present invention allows for the design of smaller ASR engines (memory consumption reduced to up to one-sixth) with reduced impact on recognition accuracy and/or robustness to noises.
45 Citations
29 Claims
-
1. A noise robust automatic speech recognition system, comprising:
-
a front end analysis module isolating a set of independent subspaces, wherein said front end analysis module employs one or more block diagonal front-end whitening matrices to isolate the set of independent subspaces; a model-compensation module employing a model-compensation distortion function that operates on each of the subspaces isolated by said front-end analysis module; and a subspace model compression module employing subspace tying to perform model compression. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
-
-
16. A method of operation for use with a noise robust automatic speech recognition system, comprising:
-
isolating a set of independent subspaces using a block diagonal front-end whitening matrix; using a model compensation module of the speech recognition system that implements a model-compensation distortion function that operates on each of the isolated subspaces; and employing subspace tying to perform model compression. - View Dependent Claims (17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29)
-
Specification