Word spotting using both filler and phone recognition

• US 5,950,159 A
• Filed: 03/19/1997
• Issued: 09/07/1999
• Est. Priority Date: 04/01/1996
• Status: Expired due to Term

• Current Assignee: HTC Corporation
• Original Assignee: Hewlett-Packard Company (HP Inc.)
• Inventors: Knill, Katherine Mary
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): SOFOCLEOUS, MICHAEL D
• Application Number:
  US08/818,646
• Time in Patent Office: 902 Days
• Field of Search: 704/253, 704/252, 704/256, 704/243, 704/251, 704/255, 704/233
• US Class Current: 704/251
• CPC Class Codes: G10L 15/04 Segmentation; Word boundary detection
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

The present invention relates to a word-spotting system and a method for finding a keyword in ascoustic data. The method includes a filler recognition phase and a keyword recognition phase wherein: during the filler recognition phase the acoustic data is processed to identify phones and to generate temporal delimiters and likelihood scores for the phones; during the keyword recognition phase, the acoustic data is processed to identify instances of a specified keyword including a sequence of phones; wherein the temporal delimiters and likelihood scores generated in the filler recognition phase are used in the keyword recognition phase.

Claims

1. A computer implemented method for finding a keyword in acoustic data, the method comprising a filler recognition phase and a keyword recognition phase that occurs subsequent to initiation of said filler recognition phase, comprising the steps of:

processing the acoustic data during the filler recognition phase to identify phones and to generate (i) temporal delimiters within said acoustic data and (ii) likelihood scores for the phones;

processing the acoustic data during the keyword recognition phase to identify instances of a specified keyword comprising a sequence of phones, said processing employing said temporal delimiters to restrict search space for instances of said specified keyword and further employing said likelihood scores generated in the filler recognition phase as an aid in the keyword recognition phase.

2. A method according to claim 1 wherein keyword recognition is performed only for portions of the acoustic data when at least one of the keyword phones is present in the related filler phone sequence.

3. A method according to claim 2 wherein said portions of the acoustic data are identified by string matching the keyword phone string against the acoustic data.

4. A method according to claim 3 wherein the string matching is performed using dynamic programming alignment.

5. The method according to claim 1, wherein said keyword recognition phase occurs subsequent to completion of said filler recognition phase.

6. A system for finding a keyword in acoustic data, the system employing a filler recognition phase and a keyword recognition phase that occurs subsequent to said initiation of said filler recognition phase and comprising:

a) means for processing the acoustic data during the filler recognition phase to identify phones and to generate a temporal delimiters for said acoustic data and likelihood scores for the phones; and

b) means for processing the acoustic data during the keyword recognition phase, using the temporal delimiters to restrict search space for instances of said keyword and likelihood scores identitfied by means a), to identify instances of said keyword comprising a sequence of phones.

7. The system according to claim 6 wherein keyword recognition is performed by means b) only for portions of the acoustic data when at least one of the keyword phones is present in the related filler phone sequence.

8. A system according to claim 7 wherein said portions of the acoustic data are identified by means b) by string matching the keyword phone string against the acoustic data.

9. A system according to claim 8 wherein the string matching is performed using dynamic programming alignment.

10. The system according to claim 9, wherein said keyword recognition phase occurs subsequent to completion of said filler recognition phase.