Acoustic model training

• US 9,697,835 B1
• Filed: 03/31/2016
• Issued: 07/04/2017
• Est. Priority Date: 03/31/2016
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Mangu, Lidia L., Thomas, Samuel, Kuo, Hong-Kwang J.
• Primary Examiner(s): Albertalli, Brian
• Application Number:
  US15/086,949
• Time in Patent Office: 460 Days
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/04 Segmentation; Word boundary detection
• CPC Class Codes: G10L 15/05 Word boundary detection
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

A method, executed by a computer, includes receiving a channel recording corresponding to a conversation, receiving a transcription for the conversation, generating a conversation-specific language model for the conversation using the transcription, and conducting speech recognition on the channel recording using the conversation-specific language model to provide time boundaries and written language corresponding to utterances within the channel recording. The method further includes determining sentence or phrase boundaries for the transcription, aligning written language within the one or more transcriptions with the written language corresponding to the utterances with the channel recording to provide sentence or phrase boundaries for the channel recording, and training a speech recognizer according to the sentence or phrase boundaries for the transcription and the sentence or phrase boundaries for the channel recording. A computer system and computer program product corresponding to the method are also disclosed herein.

Claims

1. A computer system comprising:

one or more computer processors;

one or more computer readable storage media and program instructions stored on the one or more computer readable storage media, the program instructions comprising instructions to perform;

receiving a channel recording corresponding to a conversation;

receiving a transcription for the conversation;

generating a conversation-specific language model for the conversation using the transcription;

conducting speech recognition on the channel recording using the conversation-specific language model to provide time boundaries and written language corresponding to utterances within the channel recording;

determining sentence or phrase boundaries for the transcription;

aligning written language within the one or more transcriptions with the written language corresponding to the utterances with the channel recording to provide sentence or phrase boundaries for the channel recording; and

training a speech recognizer according to the sentence or phrase boundaries for the transcription and the sentence or phrase boundaries for the channel recording.

2. The computer system of claim 1, wherein the conversation corresponds to providing customer service.

3. The computer system of claim 1, wherein conducting speech recognition on the channel recording comprises interpolating between the conversation-specific language model and a general language model.

4. The computer system of claim 1, wherein the sentence or phrase boundaries for the channel recording are determined without manual segmentation or silence-based automatic segmentation of the channel recording.

5. The computer system of claim 1, wherein training the speech recognizer comprises training an acoustic model using a channel recording and a corresponding transcript.

6. The computer system of claim 1, wherein aligning written language comprises conducting a dynamic programming procedure.

7. The computer system of claim 1, wherein the conversation-specific language models comprise an n-gram model where n is greater than or equal to 6.

8. The computer system of claim 1, further comprising trimming the channel recording according to the sentence or phrase boundaries for the channel recording.

9. A computer program product comprising:

one or more computer readable storage media and program instructions stored on the one or more computer readable storage media, the program instructions comprising instructions to perform;

receiving a channel recording corresponding to a conversation;

receiving a transcription for the conversation;

generating a conversation-specific language model for the conversation using the transcription;

conducting speech recognition on the channel recording using the conversation-specific language model to provide time boundaries and written language corresponding to utterances within the channel recording;

determining sentence or phrase boundaries for the transcription;

aligning written language within the one or more transcriptions with the written language corresponding to the utterances with the channel recording to provide sentence or phrase boundaries for the channel recording; and

training a speech recognizer according to the sentence or phrase boundaries for the transcription and the sentence or phrase boundaries for the channel recording.

10. The computer program product of claim 9, wherein the conversation corresponds to providing customer service.

11. The computer program product of claim 9, wherein conducting speech recognition on the channel recording comprises interpolating between the conversation-specific language model and a general language model.

12. The computer program product of claim 9, wherein the sentence or phrase boundaries for the channel recording are determined without manual segmentation or silence-based automatic segmentation of the channel recording.

13. The computer program product of claim 9, wherein training the speech recognizer comprises training an acoustic model using a channel recording and a corresponding transcript.

14. The computer program product of claim 9, wherein aligning written language comprises conducting a dynamic programming procedure.

15. The computer program product of claim 9, wherein the conversation-specific language models comprise an n-gram model where n is greater than or equal to 6.

16. The computer program product of claim 9, further comprising trimming the channel recording according to the sentence or phrase boundaries for the channel recording.