Automatic language model update

• US 9,159,316 B2
• Filed: 05/07/2013
• Issued: 10/13/2015
• Est. Priority Date: 04/03/2006
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Cohen, Michael H., Baluja, Shumeet, Mengibar, Pedro J.
• Primary Examiner(s): Serrou, Abdelali
• Application Number:
  US13/888,770
• Publication Number:
  US 20130246065A1
• Time in Patent Office: 889 Days
• Field of Search: 704/231, 704/244, 704/243, 704/251, 704/252, 704/253, 704/257, 704/270, 704/275
• US Class Current: 1/1
• 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 computer-implemented method, comprising:

receiving, by a computing system and as having been transmitted by a first computing device, a textual search query that was specified by user input at the first computing device, the textual search query including a textual search term;

causing, by the computing system and as a result of having received the textual search query, search results that correspond to the textual search query to be generated through a search that uses the textual search query and provided for receipt and presentation by the first computing device;

determining, by the computing system and as a result of having received the textual search query, whether the textual search term exists in a dictionary of terms for a speech recognition system;

adding, by the computing system and as a result of determining that the textual search term is not in the dictionary of terms for the speech recognition system and as a result of having received the textual search query, the textual search term to the dictionary of terms for the speech recognition system due to inclusion of the textual search term in the textual search query, in order to generate an updated dictionary of terms for the speech recognition system;

identifying, by the computing system, acoustic model data for the textual search term; and

providing, by the computing system, the acoustic model data for the textual search and the updated dictionary of terms for the speech recognition system for use in a verbal translation that includes;

receiving a verbal query that was specified by user input at another computing device and that includes a verbal representation of the textual search term, and

translating, using the acoustic model data for the textual search term and the updated dictionary of terms for the speech recognition system, the verbal query into a textual representation of the verbal query, the textual representation of the verbal query including the textual search term.

2. The computer-implemented method of claim 1, wherein determining whether the textual search term exists in the dictionary of terms for the speech recognition system includes splitting the textual search term into multiple textual search terms and determining whether each of the multiple textual search terms exists in the dictionary of terms.

3. The computer-implemented method of claim 1, further comprising:

transmitting, by the computing system, the updated dictionary of terms for the speech recognition system to the another computing device, wherein the another computing device uses the updated dictionary of terms to perform the translation of the verbal query that was received by user input specified at the another computing device into the textual representation of the verbal query.

4. The computer-implemented method of claim 1, further comprising:

assigning a probability value to a dictionary entry for the textual search term within the updated dictionary of terms based on a chronological receipt of the textual search term with respect to terms being received in other search queries.

5. The computer-implemented method of claim 1, further comprising:

assigning a probability value to a dictionary entry for the textual search term within the updated dictionary of terms based on a number of times that the textual search term is received within search queries, such that at least three dictionary entries for corresponding terms within the updated dictionary of terms have three different assigned probability values based on respective number of times that the corresponding terms have been received within the search queries.

6. The computer-implemented method of claim 1, wherein identifying the acoustic model data for the textual search term includes identifying the acoustic model data for the textual search term from a captioned video broadcast that includes captioned text that includes the textual search term.

7. The computer-implemented method of claim 1, further comprising:

adding, by the computing system and in response to receiving the textual search query, other terms that are associated with the textual search term to the dictionary of terms for the speech recognition system, wherein the other terms would not have otherwise been included in the dictionary of terms for the speech recognition system without receipt of the textual search query.

15. The computer-implemented method of claim 1, further comprising:

receiving, by the computing system and as having been transmitted by a second computing device, a second textual search query that was specified by user input at the second computing device, the second textual search query including a second textual search term;

causing, by the computing system and as a result of having received the second textual search query, search results that correspond to the second textual search query to be generated through a search that uses the second textual search query and provided for receipt and presentation by the second computing device;

assigning a first probability value to a first dictionary entry for the textual search term within the updated dictionary of terms based on a chronological receipt of the textual search term with respect to terms being received in other search queries; and

assigning a second probability value to a second dictionary entry for the second textual search term within the updated dictionary of terms based on a chronological receipt of the second textual search term with respect to terms being received in other search queries,

wherein the first probability value is different from the second probability value,

wherein the first probability value is stored concurrently with the second probability value within the updated dictionary of terms.

16. The computer-implemented method of claim 4, wherein assigning the probability value to the dictionary entry for the textual search term includes modifying an existing probability value assigned to the dictionary entry for the textual search term.

17. The computer-implemented method of claim 1, wherein:

identifying the acoustic model data for the textual search term is performed as a result of the computing system having determined that the textual search term is not in the dictionary of terms for the speech recognition system; and

providing the acoustic model data for the textual search query for use in the verbal translation is performed as a result of the computing system having determined that the textual search term is not in the dictionary of terms for the speech recognition system.

8. One or more machine-readable storage devices that are programmed with instructions that, when executed by one or more programmable processors, cause performance of operations that comprise:

receiving, by a computing system and as having been transmitted by a first computing device, a textual search query that was specified by user input at the first computing device, the textual search query including a textual search term;

causing, by the computing system and as a result of having received the textual search query, search results that correspond to the textual search query to be generated through a search that uses the textual search query and provided for receipt and presentation by the first computing device;

determining, by the computing system and as a result of having received the textual search query, whether the textual search term exists in a dictionary of terms for a speech recognition system;

adding, by the computing system and as a result of determining that the textual search term is not in the dictionary of terms for the speech recognition system and as a result of having received the textual search query, the textual search term to the dictionary of terms for the speech recognition system due to inclusion of the textual search term in the textual search query, in order to generate an updated dictionary of terms for the speech recognition system;

identifying, by the computing system, acoustic model data for the textual search term; and

providing, by the computing system, the acoustic model data for the textual search and the updated dictionary of terms for the speech recognition system for use in a verbal translation that includes;

receiving a verbal query that was specified by user input at another computing device and that includes a verbal representation of the textual search term, and

translating, using the acoustic model data for the textual search term and the updated dictionary of terms for the speech recognition system, the verbal query into a textual representation of the verbal query, the textual representation of the verbal query including the textual search term.

9. The one or more machine-readable storage devices of claim 8, wherein determining whether the textual search term exists in the dictionary of terms for the speech recognition system includes splitting the textual search term into multiple textual search terms and determining whether each of the multiple textual search terms exists in the dictionary of terms.

10. The one or more machine-readable storage devices of claim 8, wherein the operations further comprise:

transmitting, by the computing system, the updated dictionary of terms for the speech recognition system to the another computing device, wherein the another computing device uses the updated dictionary of terms to perform the translation of the verbal query that was received by user input specified at the another computing device into the textual representation of the verbal query.

11. The one or more machine-readable storage devices of claim 8, wherein the operations further comprise:

assigning a probability value to a dictionary entry for the textual search term within the updated dictionary of terms based on a chronological receipt of the textual search term with respect to terms being received in other search queries.

12. The one or more machine-readable storage devices of claim 8, wherein the operations further comprise:

assigning a probability value to a dictionary entry for the textual search term within the updated dictionary of terms based on a number of times that the textual search term is received within search queries, such that at least three dictionary entries for corresponding terms within the updated dictionary of terms have three different assigned probability values based on respective number of times that the corresponding terms have been received within the search queries.

13. The one or more machine-readable storage devices of claim 8, wherein identifying the acoustic model data for the textual search term includes identifying the acoustic model data for the textual search term from a captioned video broadcast that includes captioned text that includes the textual search term.

14. The one or more machine-readable storage devices of claim 8, wherein the operations further comprise:

adding, by the computing system and in response to receiving the textual search query, other terms that are associated with the textual search term to the dictionary of terms for the speech recognition system, wherein the other terms would not have otherwise been included in the dictionary of terms for the speech recognition system without receipt of the textual search query.