Computer-based method and apparatus for classifying statement types based on intonation analysis

• US 5,995,924 A
• Filed: 05/22/1998
• Issued: 11/30/1999
• Est. Priority Date: 05/05/1997
• Status: Expired due to Term

• Current Assignee: Qwest Communications International Incorporated (Lumen Technologies, Inc.)
• Original Assignee: MediaOne Group Inc. (Comcast Corporation), US West, Inc. (Lumen Technologies, Inc.)
• Inventors: Terry, Alvin Mark
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Lerner, Martin
• Application Number:
  US09/083,449
• Time in Patent Office: 557 Days
• Field of Search: 704/207, 704/208, 704/202, 704/226, 704/232, 704/233, 704/255
• US Class Current: 704/207
• CPC Class Codes: G10L 15/1807 using prosody or stress
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum
• CPC Class Codes: G10L 25/30 using neural networks
• CPC Class Codes: G10L 25/90 Pitch determination of speech signals

Abstract

A computer-based method and apparatus for classifying statement types using intonation analysis. The method and apparatus identify a user'"'"'s potential query when the user responds to information during dialog with an automated dialog system. Pitch information is extracted, via a cepstrum, from the speech signal. In one embodiment, the pitch intonation is processed to form a smoothed pitch or intonation contour. Then the smoothed pitch contour is processed by a set of shape detectors and this output, together with statistical information, is sent to a rule-based algorithm which attempts to classify the statement type. In another embodiment, the smoothed pitch contour is processed by a pattern recognition system such as a neural network trained with a back-propagation learning algorithm.

Claims

1. A computer-based method for classifying statement types through intonation analysis, the method comprising the steps of:

receiving a signal band limited in a region of 300 Hz to 2500 Hz which represents the energy content of a voiced utterance having a pitch in the range of 60 Hz to 600 Hz and varying background noise energy levels;

generating a signal representation of short term energy levels for the signal;

generating a background noise energy threshold level based on the signal representation;

extracting pitch values from the signal whenever the energy of the signal is above the threshold level for the entire voiced utterance;

generating a smoothed intonation contour from the pitch values;

processing segments of the smoothed intonation contour corresponding to voicing in the signal according to predetermined shape feature detectors to obtain a plurality of pitch parameters; and

applying an algorithm to the pitch parameters which algorithm distinguishes between a question statement and an acknowledgment statement.

2. The method as in claim 1 wherein the step of extracting includes the step of calculating a cepstrum from the signal whenever the energy of the signal is above the threshold level.

3. The method as in claim 2 wherein the cepstrum is calculated from the signal whenever the energy of the signal is above the threshold level by utilizing a broad time window.

4. The method as in claim 2 wherein the step of calculating is also performed whenever the energy of the signal is above a variable threshold level and segments of the signal correspond to voicing in the signal.

5. The method as in claim 1 wherein the algorithm is a rule-based algorithm.

6. The method as in claim 1 wherein the step of applying includes the step of inputting portions of the smoothed intonation contour as input vectors to a pattern recognition system to distinguish between the question statement and the acknowledgment statement.

7. The method as in claim 6 wherein the pattern recognition system includes a neural network having an input layer, a hidden layer, and an output layer.

8. The method as in claim 7 wherein the pattern recognition system performs a back-propagation analysis.

9. A computer-based apparatus for classifying statement types through intonation analysis, the apparatus comprising:

means for receiving a signal band limited in a region of 300 Hz to 2500 Hz which represents the energy content of a voiced utterance having a pitch in the range of 60 Hz to 600 Hz and varying background noise energy levels; and

a computer programmed to;

generate a signal representation of short term energy levels for the signal;

generate a background noise energy threshold level based on the signal representation;

extract pitch values from the signal whenever the energy of the signal is above the threshold level for the entire voiced utterance;

generate a smoothed intonation contour from the pitch values;

processing segments of the smoothed intonation contour corresponding to voicing in the signal according to predetermined shape feature detectors to obtain a plurality of pitch parameters; and

apply an algorithm to the pitch parameters which algorithm distinguishes between a question statement and an acknowledgment statement.

10. The apparatus as in claim 9 wherein the algorithm is a rule-based algorithm.

11. The apparatus as in claim 9 wherein portions of the smoothed intonation contour are inputted as input vectors to a pattern recognition system to distinguish between the question statement and the acknowledgment statement.

12. The apparatus as in claim 11 wherein the pattern recognition system includes a neural network having an input layer, a hidden layer, and an output layer.

13. The apparatus as in claim 12 wherein the pattern recognition system performs a back-propagation analysis.

14. The apparatus as in claim 9 wherein the cepstrum is calculated from the signal whenever the energy of the signal is above the threshold level by utilizing a broach time window.

15. The apparatus as in claim 9 wherein the programmed computer also calculates the cepstrum from the signal whenever the energy of the signal is above a variable threshold level and segments of the signal correspond to voicing in the speech signal.