Unsupervised speech model adaptation using reliable information among N-best strings
AssigneeMatsushita Electric Industrial Co., Ltd.
InventorsPatrick Nguyen, Philippe Gelin, Jean-Claude Junqua
Life of this patent
Estimating the term…
Abstract
The system performs unsupervised speech model adaptation using the recognizer to generate the N-best solutions for an input utterance. Each of these N-best solutions is tested by a reliable information extraction process. Reliable information is extracted by a weighting technique based on likelihood scores generated by the recognizer, or by a non-linear thresholding function. The system may be used in a single pass implementation or iteratively in a multi-pass implementation.
Claims26
A method for adapting a speech model based on unsupervised input speech from the user of a speech recognizer, comprising:
- using said recognizer to extract the N-best recognition solutions based on said speech model;
- selecting from said N-best recognition solutions a set of reliable recognition candidates, and
- supplying said set of reliable recognition candidates to an adaptation system to adapt said speech model.
A speech recognition apparatus that adapts an initial speech model based on unsupervised input speech from the user, comprising:
- a speech model that represents speech as a plurality of speech unit models associated with a plurality of speech units;
- a speech recognizer that processes input speech from a user using said speech model to generate the N-best recognition solutions;
- a reliable information extractor for analyzing said N-best recognition solutions to select a set of reliable recognition candidates from said N-best recognition solutions;
- and
- an adaptation system that uses said reliable recognition candidates to adapt said speech model.
Dependencies are inferred from the claim text.