Two stage utterance verification device and method thereof in speech recognition system

• US 7,529,665 B2
• Filed: 04/01/2005
• Issued: 05/05/2009
• Est. Priority Date: 12/21/2004
• Status: Expired due to Fees

• Current Assignee: Electronics and Telecommunications Research Institute
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Kim, Sanghun, Lee, YoungJik
• Primary Examiner(s): Hudspeth; David R
• Assistant Examiner(s): Rider; Justin W
• Application Number:
  US11/095,555
• Publication Number:
  US 20060136207A1
• Time in Patent Office: 1,495 Days
• Field of Search: 704/236
• US Class Current: 704/236
• CPC Class Codes: G10L 15/08 Speech classification or search

Abstract

A two stage utterance verification device and a method thereof are provided. The two stage utterance verification method includes performing a first utterance verification function based on a SVM pattern classification method by using feature data inputted from a search block of a speech recognizer and performing a second utterance verification function based on a CART pattern classification method by using heterogeneity feature data including meta data extracted from a preprocessing module, intermediate results from function blocks of the speech recognizer and the result of the first utterance verification function. Therefore, the two state utterance verification device and the method thereof provide a high quality speech recognition service to a user.

Claims

1. A method for two stage utterance verification method, comprising the steps of:

a) performing a first utterance verification function based on a support vector machine (SVM) pattern classification method by using feature data inputted from a search block of a speech recognizer;

b) determining whether a confidence score, which is a result value of the first utterance verification function, is a misrecognition level for deciding rejection of a speech recognition result;

c) performing a second utterance verification function based on a classification and regression tree (CART) pattern classification method by using heterogeneity feature data including meta data extracted from a preprocessing module, intermediate results from function blocks of the speech recognizer and the result of the first utterance verification function when the speech recognition result is accepted by the first utterance verification function, and returning when the speech recognition result is rejected by the first utterance verification function; and

d) determining whether the speech recognition result is misrecognition based on a result of the second utterance verification function, transferring the speech recognition result to a system response module when the speech recognition result is accepted by the second utterance verification, and returning when the speech recognition result is rejected by the second utterance verification.

2. The method of claim 1, wherein the heterogeneity feature data of the step c) includes a SNR, an energy, a gender, an age, a phonetic structure, a dialect, the number of syllables in a word, the number of phonemes in a word, the number of frames in a word, a speaking rate, an average pitch, an utterance duration, a speech absent probability, a speech/non-speech likelihood, a Kalman shrinking factor, a Wiener shrinking factor, a N-best LLR score, an anti-model LLR score, a filter bank SNR, a LLR driven score, a SVM confidence score, a beam width during searching, a search time, a EPD time, a time for using system and a domain.

3. A two stage utterance verification device in a speech recognition system, the device comprising:

a speech input/output unit for inputting/outputting speech;

a preprocessing module for receiving the speech from the speech input/output unit and extracting meta data from the received speech;

a speech recognizer for performing speech recognition after receiving the meta data from the preprocessing module; and

a utterance verification unit for performing a first utterance verification function based on feature data inputted from a search block of the speech recognizer by using a support vector machine (SVM) pattern classification method, performing a second utterance verification function based on a classification and regression tree (CART) pattern classification method by using heterogeneity feature data including the meta data extracted from the preprocessing module, intermediate result values from functional blocks of the speech recognizer and the result of the first utterance verification when a confidence score, which is a result of the first utterance verification function, is accepted as a correct recognition, and transferring the speech recognition result to a system response module when a result of the second utterance verification function is accepted as the correct recognition.

4. The two stage utterance verification device of claim 3, wherein the heterogeneity feature data includes a SNR, an energy, a gender, an age, a phonetic structure, a dialect, the number of syllables in a word, the number of phonemes in a word, the number of frames in a word, a speaking rate, an average pitch, an utterance duration, a speech absent probability, a speech/non-speech likelihood, a Kalman shrinking factor, a Wiener shrinking factor, a N-best LLR score, an anti-model LLR score, a filter bank SNR, a LLR driven score, a SVM confidence score, a beam width during searching, a search time, a EPD time, a time for using system and a domain.

5. A method for two stage utterance verification method, comprising the steps of:

a) performing a first utterance verification function based on a support vector machine (SVM) pattern classification method by using feature data inputted from a search block of a speech recognizer;

b) determining whether a confidence score, which is a result value of the first utterance verification function, is a misrecognition level for deciding rejection of a speech recognition result;

c) performing a second utterance verification function based on a classification and regression tree (CART) pattern classification method by using heterogeneity feature data including meta data extracted from a preprocessing module, intermediate results from function blocks of the speech recognizer and the result of the first utterance verification function when the speech recognition result is accepted by the first utterance verification function, and returning when the speech recognition result is rejected by the first utterance verification function; and

d) determining whether the speech recognition result is misrecognition based on a result of the second utterance verification function, transferring the speech recognition result to a system response module when the speech recognition result is accepted by the second utterance verification, and returning when the speech recognition result is rejected by the second utterance verification;

wherein the heterogeneity feature data of step c) includes a SNR, an energy, a gender, an age, a phonetic structure, a dialect, the number of syllables in a word, the number of phonemes in a word, the number of frames in a word, a speaking rate, an average pitch, an utterance duration, a speech absent probability, a speech/non-speech likelihood, a Kalman shrinking factor, a Wiener shrinking factor, a N-best LLR score, an anti-model LLR score, a filter bank SNR, a LLR driven score, a SVM confidence score, a beam width during searching, a search time, a EPD time, a time for using system and a domain.

6. A two stage utterance verification device in a speech recognition system, the device comprising:

a speech input/output unit for inputting/outputting speech;

a preprocessing module for receiving the speech from the speech input/output unit and extracting meta data from the received speech;

a speech recognizer for performing speech recognition after receiving the meta data from the preprocessing module; and

a utterance verification unit for performing a first utterance verification function based on feature data inputted from a search block of the speech recognizer by using a support vector machine (SVM) pattern classification method, performing a second utterance verification function based on a classification and regression tree (CART) pattern classification method by using heterogeneity feature data including the meta data extracted from the preprocessing module, intermediate result values from functional blocks of the speech recognizer and the result of the first utterance verification when a confidence score, which is a result of the first utterance verification function, is accepted as a correct recognition, and transferring the speech recognition result to a system response module when a result of the second utterance verification function is accepted as the correct recognition;

wherein the heterogeneity feature data includes a SNR, an energy, a gender, an age, a phonetic structure, a dialect, the number of syllables in a word, the number of phonemes in a word, the number of frames in a word, a speaking rate, an average pitch, an utterance duration, a speech absent probability, a speech/non-speech likelihood, a Kalman shrinking factor, a Wiener shrinking factor, a N-best LLR score, an anti-model LLR score, a filter bank SNR, a LLR driven score, a SVM confidence score, a beam width during searching, a search time, a EPD time, a time for using system and a domain.