A SYSTEM AND METHOD FOR PROVIDING LARGE VOCABULARY SPEECH PROCESSING BASED ON FIXED-POINT ARITHMETIC

• US 20070192104A1
• Filed: 02/16/2006
• Published: 08/16/2007
• Est. Priority Date: 02/16/2006
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: BLEWETT, CHARLES, BOCCHIERI, ENRICO
• Granted Patent: US 8,195,462 B2
• Time in Patent Office: Days
• US Class Current: 704/256
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/28 Constructional details of speech recognition systems

Abstract

Disclosed herein is a system, method and computer-readable medium storing instructions for controlling a computing device according to the method. The invention relates to a system, method and computer-readable medium storing instructions for controlling a computing device according to the method. As an example embodiment, the method uses a speech recognition decoder that operates or uses fixed point arithmetic. The exemplary method comprises representing arc costs associated with at least one finite state transducer (FST) in fixed point, representing parameters associated with a hidden Markov model (HMM) in fixed point and processing speech data in the speech recognition decoder using fixed point arithmetic for the fixed point FST arc costs and the fixed point HMM parameters. The method may also include computing at the decoder sentence hypothesis probabilities with fixed point arithmetic as type Q-2e numbers.

Claims

1. A method of processing data associated with speech processing, the method using a speech recognition decoder using fixed-point arithmetic, the method comprising:

representing arc costs associated with at least one finite-state transducer (FST) in fixed-point;

representing parameters associated with a hidden Markov model (HMM in fixed-point; and

processing speech data in the speech recognition decoder using fixed-point arithmetic for the fixed-point FST arc costs and the fixed-point HMM parameters.

2. The method of claim 1, wherein the method is applied for large vocabulary automatic speech recognition.

3. The method of claim 1, wherein the HMM is a phonetic context-dependent acoustic HMM.

4. The method of claim 1, wherein the at least one FST is one of a single CLG FST or distinct C, L and G FSTs.

5. The method of claim 1, wherein representing arc costs associated with at least one finite-state transducer (FST) in fixed-point further comprises:

converting floating-point costs of the at least one FST to fixed-point by quantization.

6. The method of claim 5, wherein a format for the fixed-point numbers is type Q-2e.

7. The method of claim 1, wherein representing parameters associated with a hidden Markov model (HMM in fixed-point further comprises:

converting HMM state observation probability density functions to fixed-point format by quantization; and

computing HMM state likelihoods as Q-2e fixed-point numbers.

8. The method of claim 1, wherein representing parameters associated with a hidden Markov model (HMM in fixed-point further comprises:

converting HMM state duration probabilities to fixed-point by quantization using a fixed-point format of type Q-2e.

9. The method of claim 1, further comprising:

computing at the decoder sentence hypothesis probabilities with fixed-point arithmetic as type Q-2e numbers.

10. A computing device that processes data associated with speech processing, the device using a speech recognition decoder using fixed-point arithmetic, the computing device comprising:

a module configured to represent arc costs associated with at least one finite-state transducer (FST) in fixed-point;

a module configured to represent parameters associated with a hidden Markov model (HMM) in fixed-point; and

a module configured to process speech data in the speech recognition decoder using fixed-point arithmetic for the fixed-point FST arc costs and the fixed-point HMM parameters.

11. The computing device of claim 10, wherein the computing device performs large vocabulary automatic speech recognition.

12. The computing device of claim 10, wherein the HMM is a phonetic context-dependent acoustic HMM.

13. The computing device of claim 10, wherein the at least one FST is one of a single CLG FST or distinct C, L and G FSTs.

14. The computing device of claim 10, wherein the module configured to represent arc costs associated with at least one finite-state transducer (FST) in fixed-point further converts floating-point costs of the at least one FST to fixed-point by quantization.

15. The computing device of claim 14, wherein a format for the fixed-point numbers is type Q-2e.

16. The computing device of claim 10, wherein the module configured to represent parameters associated with a hidden Markov model (HMM) in fixed-point further:

converts HMM state observation probability density functions to fixed-point format by quantization; and

computes HMM state likelihoods as Q-2e fixed-point numbers.

17. The computing device of claim 10, wherein the module configured to represent parameters associated with a hidden Markov model (HMM) in fixed-point further:

converts HMM state duration probabilities to fixed-point by quantization using a fixed-point format of type Q-2e.

18. The computing device of claim 10, further comprising:

a module configured to compute at the decoder sentence hypothesis probabilities with fixed-point arithmetic as type Q-2e numbers.

19. A computer-readable medium storing instructions for controlling a computing device to process data associated with speech processing, the computing device including a speech recognition decoder using fixed-point arithmetic, the instructions comprising:

representing arc costs associated with at least one finite-state transducer (FST) in fixed-point;

representing parameters associated with a hidden Markov model (HMM in fixed-point; and

processing speech data in the speech recognition decoder using fixed-point arithmetic for the fixed-point FST arc costs and the fixed-point HMM parameters.

20. The computer-readable medium of claim 19, wherein the step of representing arc costs associated with at least one finite-state transducer (FST) in fixed-point further comprises converting floating-point costs of the at least one FST to fixed-point by quantization.