System and method for generating a phrase pronunciation
View Patent ↗A system and method for a speech recognition technology that allows language models to be customized through the addition of special pronunciations for components of phrases, which are added to the factory language models during customization. It allows components of a phrase to have different pronunciations inside customer-added phrases than are specified for those isolated components in the factory language models.
1. A method in a computer system for adding phrase pronunciations to a language model, the method comprising steps of:
receiving at least one phrase to be added to the language model, the at least one phrase comprising a first phrase, the first phrase comprising a plurality of tokens including a first token;
generating, using the computer system, a phrase pronunciation for the first phrase comprising a token pronunciation for the first token in the first phrase, wherein generating the phrase pronunciation for the first phrase comprises determining if the first token is represented in a pron component list, and, if so, selecting as the token pronunciation for the first token in the first phrase a component pronunciation from the pron component list, wherein the pron component list comprises a list of one or more component pronunciations for at least the first token as pronounced in one or more phrases, wherein the list of one or more component pronunciations is different from any list of one or more language model pronunciations in the language model for the first token; and
adding the phrase pronunciation for the first phrase to the language model;
wherein the step of generating the phrase pronunciation for the first phrase further comprises;
if the first token is not represented in the pron component list, determining if the first token is represented in the language model, and, if so, selecting a language model pronunciation from the language model as the token pronunciation for the first token in the first phrase.
2. A method for adding phrase pronunciations to a language model, in accordance with claim 1 , wherein the pron component list includes punctuation and/or formatting that is present in the first phrase but is silent in the phrase pronunciation for the first phrase.
3. A method for adding phrase pronunciations to a language model, in accordance with claim 1 , wherein the pron component list is selected from a plurality of lists in accordance with the position of the first token within the first phrase.
4. A method for adding phrase pronunciations to a language model, in accordance with claim 1 , wherein the first token is parsed from the first phrase based on word boundaries.
5. A method for adding phrase pronunciations to a language model, in accordance with claim 4 , wherein the word boundaries comprise white spaces and/or punctuation.
6. A method for adding phrase pronunciations to a language model, in accordance with claim 1 , wherein the first token is parsed from the first phrase by looking for the longest match in the language model or a background dictionary.
7. A method for adding phrase pronunciations to a language model, in accordance with claim 1 , wherein the pron component list is one of an initial pron component list or a non-initial pron component list.
8. The method of claim 1 , wherein the list of one or more component pronunciations is different from any list of one or more language model pronunciations in the language model for the first token in that a list of one or more language model pronunciations for the first token includes a pronunciation that is not included in the list of one or more component pronunciations for the first token.
9. The method of claim 1 , wherein the list of one or more component pronunciations is different from any list of one or more language model pronunciations in the language model for the first token in that the list of one or more component pronunciations for the first token includes a pronunciation that is not included in any list of one or more language model pronunciations for the first token.
10. A computer system comprising:
a tokenizer that parses a phrase to be added to a language model into a plurality of tokens including a first token; and
a computer code mechanism that:
generates a phrase pronunciation for the phrase comprising a token pronunciation for the first token in the phrase, wherein generating the phrase pronunciation for the phrase comprises determining if the first token is represented in a pron component list, and, if so, selecting as the token pronunciation for the first token in the phrase a component pronunciation from the pron component list, wherein the pron component list comprises a list of one or more component pronunciations for at least the first token as pronounced in one or more phrases, wherein the list of one or more component pronunciations is different from any list of one or more language model pronunciations in the language model for the first token; and
adds the phrase pronunciation for the phrase to the language model;
wherein the computer code mechanism generates the phrase pronunciation for the phrase at least in party by, if the first token is not represented in the pron component list, determining if the first token is represented in the language model, and, if so, selecting a language model pronunciation from the language model as the token pronunciation for the first token in the phrase; and
wherein the tokenizer and/or the computer code mechanism is implemented by a computer.
11. The computer system of claim 10 , wherein the pron component list includes punctuation and/or formatting that is present in the phrase but is silent in the phrase pronunciation for the phrase.
12. The computer system of claim 10 , wherein the computer code mechanism selects the pron component list from a plurality of lists in accordance with the position of the first token within the phrase.
13. The computer system of claim 10 , wherein the tokenizer parses the first token from the phrase based on word boundaries.
14. The computer system of claim 13 , wherein the word boundaries comprise white spaces and/or punctuation.
15. The computer system of claim 10 , wherein the tokenizer parses the first token from the phrase by looking for the longest match in the language model or a background dictionary.
16. The computer system of claim 10 , wherein the pron component list is one of an initial pron component list or a non-initial pron component list.
17. The computer system of claim 10 , wherein the list of one or more component pronunciations is different from any list of one or more language model pronunciations in the language model for the first token in that a list of one or more language model pronunciations for the first token includes a pronunciation that is not included in the list of one or more component pronunciations for the first token.
18. The computer system of claim 10 , wherein the list of one or more component pronunciations is different from any list of one or more language model pronunciations in the language model for the first token in that the list of one or more component pronunciations for the first token includes a pronunciation that is not included in any list of one or more language model pronunciations for the first token.