Method, apparatus, and medium for measuring confidence about speech recognition in speech recognizer

• US 20070185712A1
• Filed: 06/30/2006
• Published: 08/09/2007
• Est. Priority Date: 02/09/2006
• Status: Abandoned Application

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Jeong, Jae-Hoon, Oh, Kwang Cheol
• Application Number:
  US11/477,628
• Publication Number:
  US 20070185712A1
• Time in Patent Office: Days
• US Class Current: 704/238
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems

Abstract

A method of measuring confidence of speech recognition in a speech recognizer compares a phase change point with a phoneme string change point and uses a difference between the phase change point and the phoneme string change point and a likelihood ratio, and an apparatus using the method is provided. That is, the method of the present invention includes detecting a phase change point of a speech signal; detecting a phoneme string change point according to a result of speech recognition; calculating confidence of the speech recognition by using a difference between the detected phase change point and phoneme string change point. According to the present invention, a performance of measuring confidence may become improved by simultaneously taking not only a likelihood ratio, but also taking a comparison result of a phase change point with a phoneme string change point into consideration.

Claims

1. A method of measuring confidence of speech recognition in a speech recognizer, the method comprising:

detecting a phase change point of a speech signal;

detecting a phoneme string change point according to a result of speech recognition of the speech signal; and

calculating confidence of the speech recognition by using a difference between the detected phase change point and the detected phoneme string change point, and a likelihood ratio.

2. The method of claim 1, wherein the detecting a phase change point of a speech signal detects the phase change point of the speech signal from one of a spectrogram, a waveform, and a feature of the speech signal.

3. The method of claim 2, wherein the detecting a phase change point of a speech signal comprising:

calculating a Euclidian distance between a pair of frames in the spectrogram for the speech signal; and

detecting the phase change point for the speech signal by using a calculated peak and a valley.

4. The method of claim 3, wherein the detecting a phase change point of the speech signal comprises detecting the phase change point of the speech signal by using the N-topper points of which calculated distance between the peak and the valley are higher than other points.

5. The method of claim 4, wherein the calculating confidence of the speech recognition locates an unmatched point with respect to the detected phoneme string change point among the N-topper points and calculates the confidence of the speech recognition by giving a penalty score to the unmatched point.

6. The method of claim 1, wherein the calculating confidence of the speech recognition calculates the confidence of the speech recognition by using a phase change score according to the difference and the likelihood ratio of the speech recognition.

11. A computer readable storage medium storing a program for implementing the method of claim 1.

7. A method of measuring confidence of speech recognition of a speech recognizer, the method comprising:

extracting a feature of a speech signal;

calculating a spectrogram of the speech signal;

recognizing a speech from the extracted feature of the speech signal by using a predetermined speech recognition model;

comparing a phase change of the speech signal by using a result of speech recognition and the calculated spectrogram;

calculating a likelihood ratio of the speech recognition according to the speech recognition model; and

calculating confidence of the speech recognition by considering the phase change comparison and the likelihood ratio.

8. The method of claim 7, wherein the speech recognition unit recognizes the speech through a keyword model and a filler model from the extracted feature.

9. The method of claim 8, the comparing a phase change of the speech signal by using the result of speech recognition and the calculated spectrogram comprising:

comparing a phoneme string change point which is a result of speech recognition by the keyword model with the closest phase change point of the spectrogram within a predetermined range; and

giving a penalty score to an unmatched point with respect to the phoneme string change point among N-topper points of which distance is longer than the other points according to the comparison result.

10. The method of claim 8, wherein the method further determines whether to accept the recognized speech signal or not according to the calculated confidence.

20. At least one computer readable medium comprising computer readable instructions implementing the method of claim 7.

12. A measuring apparatus for confidence of speech recognition in a speech recognizer, the apparatus comprising:

a phase change detection unit detecting a phase change point of a speech signal;

a phoneme string change detection unit detecting a phoneme string change point according to a result of speech recognition in the speech recognizer; and

a confidence calculation unit calculating confidence of the speech recognition by using a comparison result a detected phase change point with the detected phoneme string change point, and a likelihood ratio.

13. The apparatus of claim 12, wherein the phase change detection unit detects a phase change point of the speech signal from a spectrogram and a waveform of the speech signal and a feature of the speech signal.

14. The apparatus of claim 13, wherein the phase change detection unit detects a phase change point of the speech signal on a spectrogram of the speech signal by using a calculated peak and a valley.

15. The apparatus of claim 12, wherein the confidence calculation unit calculates the confidence by giving penalty scores when the detected phase change point in the spectrogram is not matched to the detected phoneme string change point

16. A measuring apparatus of confidence of speech recognition in a speech recognizer, the apparatus comprising:

a feature extraction unit extracting a feature of a speech signal;

a spectrogram calculation unit calculating a spectrogram of the speech signal;

a speech recognition unit recognizing a speech from a feature of the extracted speech signal by using a predetermined speech recognition model;

a phase change comparison unit comparing phase changes of a speech signal by using a result of speech recognition and the calculated spectrogram;

a likelihood ratio calculation unit calculating a likelihood ratio of the speech recognition according to the result of speech recognition; and

a confidence measuring unit calculating confidence of the speech recognition by considering both the comparison result of the phase change and the likelihood ratio.

17. The apparatus of claim 16, wherein the speech recognition unit recognizes the speech through a keyword model and a filler model from the extracted feature.

18. The apparatus of claim 17, wherein the phase change comparison unit comprises:

comparing a phoneme string change point which is a result of speech recognition by the keyword model with the closest point of the phase change of the spectrogram within a predetermined range; and

giving a penalty score to an unmatched point with respect to the phoneme string change point among N-topper points of which distance is longer than other points according to the comparison result.

19. The apparatus of claim 16, wherein the method further comprises a determination unit determining whether to accept the recognized speech signal or not according to the calculated confidence.

21. A method of measuring confidence of speech recognition of a speech signal comprising calculating confidence of the speech recognition by using a difference between a phase change point of the speech signal and a phoneme string change point, and by using a likelihood ratio of the speech signal.

22. At least one computer readable medium comprising computer readable instructions implementing the method of claim 21.