Word generation for speech recognition
AssigneeAmazon Technologies, Inc.
InventorJeffrey Paul Lilly
Life of this patent
Estimating the term…
Abstract
An automatic speech recognition (ASR) system may add new words to an ASR system by identifying words with similar usage and replicating the variations of the identified words to create new words. A new word that is used similarly to a known word may be varied to create new word forms that are similar to the word forms of a known word. The new word forms may then be incorporated into an ASR model to allow the ASR system to recognize those words when they are detected in speech. Such a system may allow flexible incorporation and recognition of varied forms of new words entering a general lexicon.
Claims20
A computer-implemented method for configuring a speech recognition system to recognize a new word, the method comprising:
- identifying a first speech recognition model;
- determining a first word, wherein the first speech recognition model is not configured to recognize the first word;
- determining first characteristics corresponding to how the first word is used;
- determining second characteristics corresponding to how a second word is used;
- determining that the first characteristics are similar to the second characteristics;
- determining a variation of the second word, wherein the variation of the second word is a combination of a root of the second word and a first affix;
- determining that the first speech recognition model is configured to recognize both the second word and the variation of the second word;
- determining, based at least in part on determining that the first characteristics are similar to the second characteristics and based at least in part on determining that the first speech recognition model is configured to recognize both the second word and the variation of the second word, a variation of the first word, wherein the variation of the first word is a combination of a root of the first word and the first affix;
- and
- training a second speech recognition model to recognize the first word and the variation of the first word.
A computer-implemented method comprising:
- determining a speech recognition model, wherein the speech recognition model was trained by:
- determining first characteristics corresponding to a first word,
- determining second characteristics corresponding to a second word,
- determining that the first characteristics are similar to the second characteristics,
- identifying, in a lexicon, a variation of the second word, wherein the variation of the second word differs from the second word by at least one letter,
- creating, based at least in part on determining that the first characteristics are similar to the second characteristics and based at least in part on identifying the variation of the second word, a variation of the first word, wherein the variation of the first word differs from the first word by the at least one letter, and
- training the speech recognition model to recognize the first word and the variation of the first word;
- and
- performing speech recognition using the speech recognition model to determine speech recognition output including the variation of the first word.
A computing system comprising:
- at least one processor;
- a memory including instructions operable to be executed by the at least one processor to cause the system to perform:
- determining a speech recognition model, wherein the speech recognition model was trained by:
- determining first characteristics corresponding to a first word,
- determining second characteristics corresponding to a second word,
- determining that the first characteristics are similar to the second characteristics,
- identifying, in a lexicon, a variation of the second word, wherein the variation of the second word differs from the second word by at least one letter,
- creating, based at least in part on determining that the first characteristics are similar to the second characteristics and based at least in part on identifying the variation of the second word, a variation of the first word, wherein the variation of the first word differs from the first word by the at least one letter, and
- training the speech recognition model to recognize the first word and the variation of the first word;
- and
- performing speech recognition using the speech recognition model to determine speech recognition output including the variation of the first word.
Dependencies are inferred from the claim text.