Methods and apparatus for rapid acoustic unit selection from a large speech corpus

• US 8,315,872 B2
• Filed: 11/29/2011
• Issued: 11/20/2012
• Est. Priority Date: 04/30/1999
• Status: Expired due to Fees

• Current Assignee: Cerence Operating Company (Cerence Inc.)
• Original Assignee: AT&T Intellectual Property II LP (AT&T, Inc.)
• Inventors: Riley, Michael Dennis, Beutnagel, Mark Charles, Mohri, Mehryar
• Primary Examiner(s): Lerner, Martin
• Application Number:
  US13/306,157
• Publication Number:
  US 20120136663A1
• Time in Patent Office: 357 Days
• Field of Search: 704/258, 704/260, 704/263, 704/266
• US Class Current: 704/258
• CPC Class Codes: G10L 13/00 Speech synthesis; Text to speech systems
• CPC Class Codes: G10L 13/027 Concept to speech synthesisers; Generation of natural phrases from machine-based concepts generation of parameters for speech synthesis out of text G10L13/08
• CPC Class Codes: G10L 13/07 Concatenation rules
• CPC Class Codes: G10L 13/08 Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination

Abstract

A speech synthesis system can select recorded speech fragments, or acoustic units, from a very large database of acoustic units to produce artificial speech. The selected acoustic units are chosen to minimize a combination of target and concatenation costs for a given sentence. However, as concatenation costs, which are measures of the mismatch between sequential pairs of acoustic units, are expensive to compute, processing can be greatly reduced by pre-computing and caching the concatenation costs. Unfortunately, the number of possible sequential pairs of acoustic units makes such caching prohibitive. However, statistical experiments reveal that while about 85% of the acoustic units are typically used in common speech, less than 1% of the possible sequential pairs of acoustic units occur in practice. A method for constructing an efficient concatenation cost database is provided by synthesizing a large body of speech, identifying the acoustic unit sequential pairs generated and their respective concatenation costs, and storing those concatenation costs likely to occur. By constructing a concatenation cost database in this faction, the processing power required at run-time is greatly reduced with negligible effect on speech quality.

Claims

1. A method comprising:

generating a concatenation cost database by synthesizing, via a processor, a body of speech and identifying acoustic unit sequential pairs in the body of speech and respective concatenation costs;

determining whether an acoustic unit sequential pair to be used for synthesizing speech has a concatenation cost in the concatenation cost database; and

if the concatenation cost database does not contain the concatenation cost for the acoustic unit sequential pair, calculating an actual concatenation cost for the acoustic unit sequential pair.

2. The method of claim 1, wherein if the concatenation cost database contains the concatenation cost for the acoustic unit sequential pair, then synthesizing the speech using a respective concatenation cost for the speech from the concatenation cost database.

3. The method of claim 1, further comprising synthesizing the speech using the actual concatenation cost calculated for the acoustic unit sequential pair.

4. The method of claim 1, wherein the concatenation cost database is generated by assigning costs to the acoustic unit sequential pairs.

5. The method of claim 1, wherein the concatenation cost database contains a portion of all possible concatenation costs.

6. The method of claim 1, wherein the concatenation cost database is generated using statistical techniques which predict which of the acoustic unit sequential pairs are most likely to occur in common speech.

7. The method of claim 1, wherein the actual concatenation cost comprises a weighted sum of subcosts across phones.

8. The method of claim 1, wherein the actual concatenation cost provides an estimate of an acoustic mismatch between units in the acoustic unit sequential pair.

9. A system comprising:

a processor; and

a computer readable storage medium storing instructions for controlling the processor to perform steps comprising;

generating a concatenation cost database by synthesizing, via a processor, a body of speech and identifying acoustic unit sequential pairs in the body of speech and respective concatenation costs;

determining whether an acoustic unit sequential pair to be used for synthesizing speech has a concatenation cost in the concatenation cost database; and

if the concatenation cost database does not contain the concatenation cost for the acoustic unit sequential pair, calculating an actual concatenation cost for the acoustic unit sequential pair.

10. The system of claim 9, further comprising if the concatenation cost database contains the concatenation cost for the acoustic unit sequential pair, synthesizing the speech using a respective concatenation cost for the speech from the concatenation cost database.

11. The system of claim 9, further comprising synthesizing the speech using the actual concatenation cost calculated for the acoustic unit sequential pair.

12. The system of claim 9, wherein the concatenation cost database contains a portion of all possible concatenation costs.

13. The system of claim 9, wherein the concatenation cost database is generated using statistical techniques which predict which of the acoustic unit sequential pairs are most likely to occur in common speech.

14. The system of claim 9, wherein the actual concatenation cost comprises a weighted sum of subcosts across phones.

15. The system of claim 9, wherein the actual concatenation cost provides an estimate of an acoustic mismatch between units in the acoustic unit sequential pair.

16. A non-transitory computer-readable storage media storing instructions which, when executed by a computing device, cause the computing device to perform steps comprising:

generating a concatenation cost database by synthesizing, via a processor, a body of speech and identifying acoustic unit sequential pairs in the body of speech and respective concatenation costs;

determining whether an acoustic unit sequential pair to be used for synthesizing speech has a concatenation cost in the concatenation cost database; and

if the concatenation cost database does not contain the concatenation cost for the acoustic unit sequential pair, calculating an actual concatenation cost for the acoustic unit sequential pair.

17. The non-transitory computer-readable storage media of claim 16, further comprising if the concatenation cost database contains the concatenation cost for the acoustic unit sequential pair, synthesizing the speech using a respective concatenation cost for the speech from the concatenation cost database.

18. The non-transitory computer-readable storage media of claim 16, further comprising synthesizing the speech using the actual concatenation cost for the acoustic unit sequential pair.

19. The non-transitory computer-readable storage media of claim 16, wherein the concatenation cost database is generated using statistical techniques which predict which of the acoustic unit sequential pairs are most likely to occur in common speech.

20. The non-transitory computer-readable storage media of claim 16, wherein the actual concatenation cost provides an estimate of an acoustic mismatch between units in the acoustic unit sequential pair.