Automatic language model update

• US 9,953,636 B2
• Filed: 10/09/2015
• Issued: 04/24/2018
• Est. Priority Date: 04/03/2006
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google LLC (Alphabet Inc.)
• Inventors: Cohen, Michael H., Baluja, Shumeet, Moreno Mengibar, Pedro J.
• Primary Examiner(s): Serrou, Abdelali
• Application Number:
  US14/879,755
• Publication Number:
  US 20160035345A1
• Time in Patent Office: 928 Days
• CPC Class Codes: G10L 15/06 Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice G10L15/14 takes precedence
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/0635 updating or merging of old and new templates; Mean values; Weighting

Abstract

A method for generating a speech recognition model includes accessing a baseline speech recognition model, obtaining information related to recent language usage from search queries, and modifying the speech recognition model to revise probabilities of a portion of a sound occurrence based on the information. The portion of a sound may include a word. Also, a method for generating a speech recognition model, includes receiving at a search engine from a remote device an audio recording and a transcript that substantially represents at least a portion of the audio recording, synchronizing the transcript with the audio recording, extracting one or more letters from the transcript and extracting the associated pronunciation of the one or more letters from the audio recording, and generating a dictionary entry in a pronunciation dictionary.

Claims

1. A method comprising:

accessing a baseline language model that associates a respective baseline probability of occurrence with each of multiple different terms;

obtaining information related to recent language usage from recent search queries that were submitted by multiple users of a search engine within a predetermined period of time;

determining a quantity of occurrences of a particular term in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time;

selectively modifying the baseline language model to independently revise the baseline probability of occurrence associated with the particular term based at least on the quantity of occurrences of the particular term in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time while maintaining unchanged a baseline probability of occurrence associated with a different term that does not occur in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time by assigning a first probability to the particular term in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time that is greater than a second probability for the different term that does not occur in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time; and

generating, by an automated speech recognizer using the modified language model, a transcription of one or more utterances of one or more different users of the search engine.

2. The method of claim 1, wherein the particular term comprises a word.

3. The method of claim 1, further comprising receiving from a remote device a verbal search term that is associated, using a recognizer implementing the language model, with a text search term.

4. The method of claim 3, further comprising transmitting to the remote device search results associated with the text search term.

5. The method of claim 3, further comprising transmitting to the remote device the text search term.

6. The method of claim 1, further comprising receiving a verbal command for an application that is associated, using a recognizer implementing the language model, with a text command.

7. The method of claim 6, wherein the language model assigns the verbal command to a sub-grammar slot.

8. The method of claim 1, wherein the language model is a rule-based model, a statistical model, or both.

9. The method of claim 1, further comprising transmitting to a remote device at least a portion of the modified language model.

10. The method of claim 9, wherein the remote device accesses the portion of the modified language model to perform speech recognition functions.

11. The method of claim 1, wherein the language model includes weightings for co-concurrence events between two or more words.

12. The method of claim 1, wherein the language model includes weightings associated with when the search queries are received.

13. The method of claim 1, wherein obtaining the information related to recent language within the predetermined period of time includes generating word counts for each word.

14. A system comprising:

a request processor to receive search terms, each of the search terms being terms in a language;

an extractor for obtaining information related to recent usage in the language from recent search terms of the search terms, the recent search terms from recent search queries that were submitted by multiple users of a search engine within a predetermined period of time;

means for selectively modifying a language model for the language that associates a respective baseline probability of occurrence with each of multiple different terms in the language, to independently revise the baseline probability occurrence associated with a particular term based at least on a quantity of occurrences of the particular term in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time while maintaining unchanged a baseline probability of occurrence associated with a different term that does not occur in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time by assigning a first probability to the particular term in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time that is greater than a second probability for the different term that does not occur in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time; and

means for generating, by an automated speech recognizer using the modified language model, a transcription of one or more utterances of one or more different users of the search engine.

17. The system of claim 14, wherein the particular term and the different term comprise terms for which the automated speech recognizer has uncertainty determining which of the terms was spoken.

15. A computer implemented method comprising:

transmitting, from a remote device to a server device, recent search terms from recent search queries that were submitted by multiple users of a search engine within a predetermined period of time, each of the recent search terms being terms in a language, wherein the server device;

generates word occurrence data associated with the recent search terms from recent search queries that were submitted by multiple users of the search engine within the predetermined period of time, the word occurrence data including at least a quantity of occurrences of a particular term in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time,

selectively modifies a baseline language model for the language that associates respective word occurrence data with each of multiple different terms in the language to independently revise the word occurrence data associated with the recent search terms from recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time, while maintaining unchanged respective word occurrence data associated with different terms that do not occur in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time by assigning a first probability to a particular term in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time that is greater than a second probability for a different term that does not occur in the recent search queries that were submitted by the multiple users of the search engine within the predetermined period of time, and

generates, by an automated speech recognizer using the modified language model, a transcription of one or more utterances of one or more different users of the search engine.

16. The computer implemented method of claim 15, wherein the remote device is selected from a group consisting of a mobile telephone, a personal digital assistant, a desktop computer, and a mobile email device.