Phonetic fragment search in speech data

• US 20050159953A1
• Filed: 01/15/2004
• Published: 07/21/2005
• Est. Priority Date: 01/15/2004
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Chang, Eric I-Chao, Seide, Frank Torsten
• Granted Patent: US 7,401,019 B2
• Time in Patent Office: Days
• US Class Current: 704/254
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

A method of searching audio data is provided including receiving a query defining multiple phonetic possibilities. The method also includes comparing the query with a lattice of phonetic hypotheses associated with the audio data to identify if at least one of the multiple phonetic possibilities is approximated by at least one phonetic hypothesis in the lattice of phonetic hypotheses.

Claims

1. A method of searching audio data, comprising:

receiving a query defining multiple phonetic possibilities; and

comparing the query with a lattice of phonetic hypotheses associated with the audio data to identify if at least one of the multiple phonetic possibilities is approximated by at least one phonetic hypothesis in the lattice of phonetic hypotheses.

2. The method of claim 1 and further comprising:

calculating a score indicative of the difference between said at least one of the multiple phonetic possibilities and said at least one hypothesis in the lattice.

4. The method of claim 1 and further comprising:

identifying a plurality of phonetic hypotheses in the lattice of phonetic hypotheses that approximate the query.

5. The method of claim 4 and further comprising:

ranking the plurality of phonetic hypotheses identified.

6. The method of claim 1 and further comprising:

identifying a time span associated with said at least one phonetic hypothesis in the lattice of phonetic hypotheses.

7. The method of claim 1 wherein the query is represented as at least one of a finite-state network, a context-free grammar and a prefix tree.

16. The computer readable medium of claim 1 wherein the data structure further includes a plurality of fragment transitions connecting one phoneme hypothesis in one phonetic fragment to another phoneme hypothesis in another phonetic fragment.

3. The method of claim 3 and further comprising:

adding a penalty value to the score if said at least one of the multiple phonetic possibilities is approximated by disconnected phonetic hypotheses.

8. A method of generating a lattice from audio data, comprising:

recognizing phonetic fragments within the audio data, wherein at least some of the phonetic fragments include at least two phones; and

calculating a score for paths joining adjacent phonetic fragments.

9. The method of claim 8 and further comprising:

calculating time values and individual acoustic scores for each phone of each of the phonetic fragments.

10. The method of claim 8 and further comprising:

pruning paths that have a score that does not meet a threshold level.

11. The method of claim 8 and further comprising:

collapsing a plurality of audio frames in the audio data into a single audio frame.

12. A computer readable-medium encoded with a data structure, comprising:

a plurality of phoneme hypotheses and an associated score for each hypothesis, wherein at least some of the hypotheses form phonetic fragments that include at least two phones; and

a plurality of transitions connecting the phoneme hypotheses.

13. The computer readable-medium of claim 12 wherein the plurality of phoneme hypotheses corresponds to recorded audio data.

14. The computer readable medium of claim 13 wherein each associated score represents the likelihood of the phoneme hypotheses given the recorded audio data.

15. The computer readable medium of claim 12 wherein the plurality of transitions include an associated time value.