Adapting a compressed model for use in speech recognition
AssigneeMicrosoft Corporation
InventorsJinyu Li, Li Deng, Dong Yu, Jian Wu, Yifan Gong, Alejandro Acero
Life of this patent
Estimating the term…
Abstract
A speech recognition system includes a receiver component that receives a distorted speech utterance. The speech recognition also includes an adaptor component that selectively adapts parameters of a compressed model used to recognize at least a portion of the distorted speech utterance, wherein the adaptor component selectively adapts the parameters of the compressed model based at least in part upon the received distorted speech utterance.
Claims20
A speech recognition system comprising the following computer-executable components:
- a receiver component that receives a distorted speech utterance;
- and
- an adaptor component in communication with the receiver component, wherein the adaptor component selectively adapts parameters of a compressed model used to recognize at least a portion of the distorted speech utterance, wherein the adaptor component selectively adapts the parameters of the compressed model based at least in part upon the received distorted speech utterance.
A method comprising the following computer-executable acts:
- receiving a distorted speech utterance at a mobile device that includes a speech recognition system;
- and
- adapting parameters of a compressed model in the speech recognition system based at least in part upon the distorted speech utterance.
A computer-readable medium residing in a mobile device that includes a speech recognition system comprising a plurality of compressed models, wherein the computer-readable medium comprises instructions that, when executed by a processor, perform the following acts:
- receive a distorted speech utterance;
- coarsely estimate an additive distortion mean vector corresponding to the distorted speech utterance based at least in part upon a portion of the received distorted speech utterance;
- adapt parameters of a first subset of compressed models in the plurality of compressed models based at least in part upon the coarsely estimated additive distortion mean vector;
- decode the received distorted speech utterance using a second subset of compressed models in the plurality of compressed models;
- re-estimate the additive distortion mean vector based at least in part upon the decoded speech utterance;
- adapt parameters of the second subset of compressed models in the plurality of compressed models based at least in part upon the re-estimated additive distortion mean vector;
- and
- decode the received distorted speech utterance using at least one compressed model in the second subset of compressed models.
Dependencies are inferred from the claim text.