Method of active learning for automatic speech recognition

• US 8,990,084 B2
• Filed: 02/10/2014
• Issued: 03/24/2015
• Est. Priority Date: 07/29/2002
• Status: Expired due to Term

• Current Assignee: Interactions, LLC
• Original Assignee: Interactions, LLC
• Inventors: Gorin, Allen Louis, Hakkani-Tur, Dilek Z., Riccardi, Guiseppe
• Primary Examiner(s): Armstrong, Angela A
• Application Number:
  US14/176,439
• Publication Number:
  US 20140156275A1
• Time in Patent Office: 407 Days
• Field of Search: 704/231, 704/235, 704/240, 704/243, 704/244, 704/251, 704/255, 704/257
• US Class Current: 704/243
• CPC Class Codes: G10L 15/063 Training

Abstract

State-of-the-art speech recognition systems are trained using transcribed utterances, preparation of which is labor-intensive and time-consuming. The present invention is an iterative method for reducing the transcription effort for training in automatic speech recognition (ASR). Active learning aims at reducing the number of training examples to be labeled by automatically processing the unlabeled examples and then selecting the most informative ones with respect to a given cost function for a human to label. The method comprises automatically estimating a confidence score for each word of the utterance and exploiting the lattice output of a speech recognizer, which was trained on a small set of transcribed data. An utterance confidence score is computed based on these word confidence scores; then the utterances are selectively sampled to be transcribed using the utterance confidence scores.

Claims

1. A method comprising:

recognizing untranscribed audio utterances that are candidates for transcription using a trained acoustic model and a trained language model;

computing confidence scores associated with an accuracy of speech recognition of the untranscribed audio utterances;

transcribing an audio utterance from the untranscribed audio utterances, the audio utterance having a lowest confidence score in the confidence scores; and

removing the audio utterance which was transcribed from a database comprising the untranscribed audio utterances.

2. The method of claim 1, further comprising iteratively repeating the transcribing with additional audio utterances from the untranscribed audio utterances until a word accuracy converges.

3. The method of claim 1, wherein the audio utterance comprises a plurality of audio works.

4. The method of claim 1, further comprising leaving out from consideration for transcription utterances with confidence scores indicating that the untranscribed audio utterances were correctly recognized.

5. The method of claim 1, wherein word posterior probability estimates are used for the confidence scores associated with the untranscribed audio utterances.

6. The method of claim 5, wherein a word is considered to be correctly recognized when an associated posterior probability is higher than a second threshold value.

7. A system comprising:

a processor; and

a computer-readable storage medium having instructions stored which, when executed by the processor, cause the processor to perform operations comprising;

recognizing untranscribed audio utterances that are candidates for transcription using a trained acoustic model and a trained language model;

computing confidence scores associated with an accuracy of speech recognition of the untranscribed audio utterances;

transcribing an audio utterance from the untranscribed audio utterances, the audio utterance having a lowest confidence score in the confidence scores; and

removing the audio utterance which was transcribed from a database comprising the untranscribed audio utterances.

8. The system of claim 7, the computer-readable storage medium having additional instructions stored which result in operations comprising iteratively repeating the transcribing with additional audio utterances from the untranscribed audio utterances until a word accuracy converges.

9. The system of claim 7, wherein the audio utterance comprises a plurality of audio works.

10. The system of claim 7, the computer-readable storage medium having additional instructions stored which result in operations comprising leaving out from consideration for transcription utterances with confidence scores indicating that the untranscribed audio utterances were correctly recognized.

11. The system of claim 7, wherein word posterior probability estimates are used for the confidence scores associated with the untranscribed audio utterances.

12. The system of claim 11, wherein a word is considered to be correctly recognized when an associated posterior probability is higher than a second threshold value.

13. A computer-readable storage device having instructions stored which, when executed by a computing device, cause the computing device to perform operations comprising:

recognizing untranscribed audio utterances that are candidates for transcription using a trained acoustic model and a trained language model;

computing confidence scores associated with an accuracy of speech recognition of the untranscribed audio utterances;

transcribing an audio utterance from the untranscribed audio utterances, the audio utterance having a lowest confidence score in the confidence scores; and

removing the audio utterance which was transcribed from a database comprising the untranscribed audio utterances.

14. The computer-readable storage device of claim 13, having additional instructions stored which result in operations comprising iteratively repeating the transcribing with additional audio utterances from the untranscribed audio utterances until a word accuracy converges.

15. The computer-readable storage device of claim 13, wherein the audio utterance comprises a plurality of audio works.

16. The computer-readable storage device of claim 13, having additional instructions stored which result in operations comprising leaving out from consideration for transcription utterances with confidence scores indicating that the untranscribed audio utterances were correctly recognized.

17. The computer-readable storage device of claim 16, wherein word posterior probability estimates are used for the confidence scores associated with the untranscribed audio utterances.