SYSTEM AND METHOD OF EXPLOITING PROSODIC FEATURES FOR DIALOG ACT TAGGING IN A DISCRIMINATIVE MODELING FRAMEWORK

• US 20090119102A1
• Filed: 11/01/2007
• Published: 05/07/2009
• Est. Priority Date: 11/01/2007
• Status: Active Grant

• Current Assignee: AT&T Labs Incorporated (AT&T, Inc.)
• Original Assignee: AT&T Labs Incorporated (AT&T, Inc.)
• Inventors: RANGARAJAN SRIDHAR, Vivek Kumar, BANGALORE, Srinivas
• Granted Patent: US 7,996,214 B2
• Time in Patent Office: Days
• US Class Current: 704/240
• CPC Class Codes: G10L 15/1807 using prosody or stress
• CPC Class Codes: G10L 25/90 Pitch determination of speech signals

Abstract

Disclosed are a system and method for exploiting information in an utterance for dialog act tagging. An exemplary method includes receiving a user utterance, computing at periodic intervals at least one parameter in the user utterance, quantizing the at least one parameter at each periodic interval, approximating conditional probabilities using an n-gram over a sliding window over the periodic intervals and tagging the utterance as a dialog act based on the approximated conditional probabilities.

Claims

1. A method for exploiting information in an utterance for dialog act tagging, the method comprising:

receiving a user utterance;

computing at periodic intervals at least one parameter in the user utterance;

quantizing the at least one parameter at each periodic interval;

approximating conditional probabilities using an n-gram over a sliding window over the periodic intervals; and

tagging the utterance as a dialog act based on the approximated conditional probabilities.

2. The method of claim 1, wherein the at least one parameter comprises a pitch and an energy.

3. The method of claim 1, wherein the conditional probability is further approximated based on a length of the utterance.

4. The method of claim 1, wherein the quantized at least one parameter at the periodic intervals over at least a portion of the utterance are used as a feature vector for a user with the n-gram model.

5. The method of claim 2, wherein the energy and pitch are normalized and only the pitch is quantized.

6. The method of claim 1, wherein approximating conditional probabilities further comprises classifying the at least one parameter using a maximum entropy discriminative classifiers.

7. The method of claim 1, wherein the at least one parameter is an acoustic-prosodic feature.

8. The method of claim 1, wherein the at least one parameter comprises at least one of acoustic-prosodic, lexical, syntactic, and syntax-based prosodic cues.

9. A system for exploiting information in an utterance for dialog act tagging, the system comprising:

a module configured to receive a user utterance;

a module configured to compute at periodic intervals at least one parameter in the user utterance;

a module configured to quantize the at least one parameter at each periodic interval;

a module configured to approximate conditional probabilities using an n-gram over a sliding window over the periodic intervals; and

a module configured to tag the utterance as a dialog act based on the approximated conditional probabilities.

10. The system of claim 9, wherein the at least one parameter comprises a pitch and an energy.

11. The system of claim 9, wherein the conditional probability is further approximated based on a length of the utterance.

12. The system of claim 9, wherein the quantized at least one parameter at the periodic intervals over at least a portion of the utterance are used as a feature vector for a user with the n-gram model.

13. The system of claim 10, wherein the energy and pitch are normalized and only the pitch is quantized.

14. The system of claim 9, wherein the module configured to approximate conditional probabilities further classifies the at least one parameter using a maximum entropy discriminative classifier.

15. The system of claim 9, wherein the at least one parameter is an acoustic-prosodic feature.

16. The system of claim 9, wherein the at least one parameter comprises at least one of acoustic-prosodic, lexical, syntactic, and syntax-based prosodic cues.

17. A computer-readable medium storing instructions for controlling a computer device to exploit information in an utterance for dialog act tagging, the instructions comprising:

receiving a user utterance;

computing at periodic intervals at least one parameter in the user utterance;

quantizing the at least one parameter at each periodic interval;

approximating conditional probabilities using an n-gram over a sliding window over the periodic intervals; and

tagging the utterance as a dialog act based on the approximated conditional probabilities.

18. The computer-readable medium of claim 17, wherein the at least one parameter comprises a pitch and an energy.

19. The computer-readable medium of claim 17, wherein the conditional probability is further approximated based on a length of the utterance.

20. The computer-readable medium of claim 17, wherein the quantized at least one parameter at the periodic intervals over at least a portion of the utterance are used as a feature vector for a user with the n-gram model.