Generating Topic-Specific Language Models

• US 20110004462A1
• Filed: 07/01/2009
• Published: 01/06/2011
• Est. Priority Date: 07/01/2009
• Status: Active Grant

• Current Assignee: TiVo Corporation (Adeia Inc.)
• Original Assignee: Comcast Interactive Media LLC (Comcast Corporation)
• Inventors: Houghton, David F., Murray, Seth Michael, Simon, Sibley Verbeck
• Granted Patent: US 9,892,730 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams

Abstract

Speech recognition may be improved by generating and using a topic specific language model. A topic specific language model may be created by performing an initial pass on an audio signal using a generic or basis language model. A speech recognition device may then determine topics relating to the audio signal based on the words identified in the initial pass and retrieve a corpus of text relating to those topics. Using the retrieved corpus of text, the speech recognition device may create a topic specific language model. In one example, the speech recognition device may adapt or otherwise modify the generic language model based on the retrieved corpus of text.

Claims

1. A method comprising:

receiving, at a speech recognition device, an audio signal;

determining a first topic of the audio signal;

retrieving a plurality of words relating to the determined first topic; and

generating, by the speech recognition device, a language model related to the determined first topic using the retrieved plurality of words.

2. The method of claim 1, wherein determining the first topic of the audio signal comprises:

identifying a plurality of words used in the audio signal, the plurality of words including a word corresponding to the first topic of the audio signal; and

determining at least one of;

a frequency of the word and whether the word is specified as being meaningful.

3. The method of claim 1, further comprising determining a second topic of the audio signal and wherein retrieving a plurality of words relating to the determined first topic includes retrieving one or more words relating to the determined second topic.

4. The method of claim 3, wherein the language model is related to the determined first and second topics.

5. The method of claim 1, wherein retrieving the plurality of words includes:

generating a search query relating to the determined first topic;

receiving a first plurality of search results corresponding to the search query; and

extracting text from the first plurality of search results.

6. The method of claim 5, further comprising removing extraneous information from the extracted text.

7. The method of claim 6, wherein the extraneous information includes formatting codes.

8. The method of claim 6, wherein the extraneous information includes metadata.

9. The method of claim 5, further comprising:

determining whether a number of words in the extracted text meets a specified threshold number of words; and

in response to determining that the number of words does not meet the specified threshold number, extracting text from a second plurality of search results.

10. The method of claim 1, further comprising recognizing one or more words from the audio signal using the generated language model.

11. The method of claim 1, wherein generating the topic specific language model includes determining a probability that a first word of the plurality of words will follow a second word of the plurality of words based on a sequence of the retrieved plurality of words.

12. The method of claim 11, wherein generating the topic specific language model includes determining a probability that a third word of the plurality of words will follow the second word based on the sequence of the retrieve plurality of words.

13. An apparatus comprising:

a processor; and

memory operatively coupled to the processor, the memory storing computer readable instructions that, when executed, cause the apparatus to;

receive, at a speech recognition device, an audio signal;

determine a first topic of the audio signal;

retrieve a plurality of words relating to the determined first topic; and

generate, by the speech recognition device, a language model related to the determined first topic using the retrieved plurality of words.

14. The apparatus of claim 13, wherein determining the first topic of the audio signal comprises:

identifying a plurality of words used in the audio signal, the plurality of words including a word corresponding to the first topic of the audio signal; and

determining at least one of;

a frequency of the word and whether the word is specified as being meaningful.

15. The apparatus of claim 13, wherein the instructions, when executed, further cause the apparatus to determine a second topic of the audio signal and wherein retrieving a plurality of words relating to the determined first topic includes retrieving one or more words relating to the determined second topic.

16. The apparatus of claim 13, wherein the language model is related to the determined first and second topics.

17. The apparatus of claim 13, wherein retrieving the plurality of words includes:

generating a search query relating to the determined first topic;

receiving a first plurality of search results corresponding to the search query; and

extracting text from the first plurality of search results.

18. The apparatus of claim 17, wherein the instructions, when executed, further cause the apparatus to remove extraneous information from the extracted text.

19. The apparatus of claim 17, wherein the instructions, when executed, further cause the apparatus to:

determine whether a number of words in the extracted text meets a specified threshold number of words; and

in response to determining that the number of words does not meet the specified threshold, extract text from a second plurality of search results.

20. One or more computer readable media storing computer readable instructions that, when executed, cause an apparatus to:

receive, at a speech recognition device, an audio signal;

determine a first topic of the audio signal;

retrieve a plurality of words relating to the determined first topic; and

generate, by the speech recognition device, a language model related to the determined first topic using the retrieved plurality of words.

21. The one or more computer readable media of claim 20, wherein determining the first topic of the audio signal includes:

identifying a plurality of words used in the audio signal, the plurality of words including a word corresponding to the first topic of the audio signal; and

determining at least one of;

a frequency of the word and whether the word is specified as being meaningful.

22. The one or more computer readable media of claim 20, wherein the instructions, when executed, further cause the apparatus to determine a second topic of the audio signal and wherein retrieving a plurality of words relating to the determined first topic includes retrieving one or more words relating to the determined second topic.

23. A speech recognition system comprising:

a processor; and

memory operatively coupled to the processor, wherein the memory stores computer readable instructions that, when executed, cause the system to;

receive, at the speech recognition system, an audio signal;

perform a first speech recognition pass on the audio signal using a first language model to identify one or more words of the audio signal;

determine a first topic of the audio signal based on the identified one or more words;

generate a search query relating to the topic of the audio signal;

retrieve a plurality of text relating to the determined first topic using the search query;

extract a plurality of words from the retrieved plurality of text;

generate, by the speech recognition device, a second language model related to the determined topic using the extracted plurality of words; and

perform a second speech recognition pass on the audio signal using the second language model specific to the determined topic.

24. The speech recognition system of claim 23, wherein the first language model includes a probability that a first word will follow a second word.

25. The speech recognition system of claim 23, wherein generating the second language model includes modifying the probability that the first word will follow the second word based on a sequence of the plurality of words in the retrieved plurality of text.