[논문]A Simple Speech/Non-speech Classifier Using Adaptive Boosting

Kwon, Oh-Wook; Lee, Te-Won

A Simple Speech/Non-speech Classifier Using Adaptive Boosting 원문보기

The journal of the Acoustical Society of Korea, v.22 no.3E, 2003년, pp.124 - 132

Kwon, Oh-Wook (School of Electrical and Computer Engineering, Chungbuk National University) , Lee, Te-Won (Institute for Neural Computation, University of California)

Abstract ▼ AI-Helper

We propose a new method for speech/non-speech classifiers based on concepts of the adaptive boosting (AdaBoost) algorithm in order to detect speech for robust speech recognition. The method uses a combination of simple base classifiers through the AdaBoost algorithm and a set of optimized speech features combined with spectral subtraction. The key benefits of this method are the simple implementation, low computational complexity and the avoidance of the over-fitting problem. We checked the validity of the method by comparing its performance with the speech/non-speech classifier used in a standard voice activity detector. For speech recognition purpose, additional performance improvements were achieved by the adoption of new features including speech band energies and MFCC-based spectral distortion. For the same false alarm rate, the method reduced 20-50% of miss errors.

주제어

AI 본문요약
AI-Helper

* AI 자동 식별 결과로 적합하지 않은 문장이 있을 수 있으니, 이용에 유의하시기 바랍니다.

문제 정의

For that purpose, a decision tree was used to combine multiple features fbr in中roved accuracy[4]. The purpose of this study was to design improved speech detectors robust to noise in various environments. The speech detectors should be designed automatically fi'om data to relieve human efforts.

제안 방법

729 VAD with comparable computational complexity. After confirming the validity of the AdaBoost algorithm we proposed a new speech speech/non-speech classifier fbr speech recognition by using new features and weights for spectral-subtracted signals. Further analysis of the learned weights for the base classifiers revealed the contribution of each feature component.
After confirming the validity of the AdaBoost algorithm we proposed a new speech speech/non-speech classifier for speech recognition by using new features and weights for spectral-subtracted signals.
For speech detection purposes we attempted to use better features derived from feature extraction for speech recognition as described in Section 22 As shown in Figure 7, we analyzed the learned weight values and found that the low- and pass-band energies contribute to the performance of the classifier compared with the G.729-based features.
For speech signals in babble noise environments, the method reduced 20-50% of miss errors for the same false alarm rate. Our proposed method gives good classification accuracy by combining simple linear base classifiers. V.
The proposed algorithm is intended for robust speech recognition.

이론/모형

In our design we can keep the simplicity of the speech/non-speech classifier of the ITU-T Recommendation G.729 Annex B while we obtain the hyperplanes in a principled and automatic manner by using the AdaBoost algorithm. The AdaBoost algorithm was chosen among many classifiers due to its simple use of base classifiers and its non-overfitting property.
Speech/Non-speech classifier using the AdaBoost algorithm.
The speech detectors should be designed automatically fi'om data to relieve human efforts. The computation load should be as low as the G.729 voice activity detector (VAD) which uses multiple decision hyperplanes fbr s peech/non-speech classification. To satisfy the constraints, we apply an adaptive boosting (AdaBoost) algorithm[9,10] to speech/non-speech classifier design.
The computation load should be as low as the G.729 voice activity detector (VAD) which uses multiple decision hyperplanes for s peech/non-speech classification.

성능/효과

IV. Conclusions We proposed a new speech/non-speech classification algorithm based on the AdaBoost algorithm and a set of new speech features. Our experimental results indicate that a nearly optimal classifier can be designed automatically with computational con屮lexity coirq)arable to the manuallyoptimized classifier.
With the decision stump, the base classifiers have axis-parallel decision boundaries and the final classifier used in speech detection can be implemented by comparing with threshold values determined by the parameters of the base classifiers. Experimental results showed that performance difference between the decision stump and the perceptron as the base classifier was not significant. In both cases, the classifiers converged sufficiently with 100 base classifiers and the difference between the training data set and the validation data set was small.
Conclusions We proposed a new speech/non-speech classification algorithm based on the AdaBoost algorithm and a set of new speech features. Our experimental results indicate that a nearly optimal classifier can be designed automatically with computational con屮lexity coirq)arable to the manuallyoptimized classifier. We investigated the contribution of each feature for speech/non-speech classification and the effects of spectral subtraction.

후속연구

On the contrary, the decision stump-based speech/ non-speech classifier requires 1 addition of the bias term, 1 table lookup for tanh( ) and 1 multiplication by the weight for each base 이assifier. The required computation is linearly proportional to the number of base classifiers, which can be tuned with additional experiments. IV.

참고문헌 (13)

H. K. Kim, and R. C. Cox, 'Evaluation of robust speech recognition algorithms for distributed speech recognition in a noisy automobile environment,' Proc. ICSLP 2002, 233-236, Sept. 2002
C.-P. Chen, K. Filali, amd J. F. Bilmes, 'Frontend postprocessing and backend model enhancement on the Aurora 2.0/3.0 databases,' Proc. ICSLP 2002, 241-244, Sept. 2002
M. Marzinzik, and B. Kollmeier, 'Speech pause detection for noise spectrum estimation by tracking power envelope dynamics,' IEEE Trans. Speech and Audio Processing, 10 (2). 109-110, Feb. 2002

상세보기
W.-H. Shin, B.-S. Lee, Y.-K. Lee. and J.-S. Lee, 'Speech/Non-speech classification using multiple features for robust endpoint detection'" Proc. ICASSP 2000. Ill. 1399-IIl. 1402, 2000
B. Kotnik, Z. Kacic, and B. Horvat, 'A computational efficient real time noise robust speech recognition based on improved spectral subtraction method.' Proc. EUROSPEECH 2001, 1123-1126, 2001
L. Karray, and A. Martin, 'Towards improving speech detection robustness for speech recognition in adverse conditions'" Speech Communication. 40 (3), 261-276, May 2003

상세보기
R. Martin, 'Spectral subtraction based on minimum statistics.' Signal Processing VII, Theories and Applications. Proc. EUSIPCO-94, 1182-1185. 1994
N. W. D. Evans, and J. S. Mason, 'Noise estimation without explicit speech, non-speech detection: A comparison of mean, modal and median based approaches,' Proc. EUROSPEECH 2001. 893-896, 2001
R. E. Schapire, and Y. Singer, "Improved boosting algorithms using confidence -rated predictions,' Machine Learning, 37 (3), 297-336, 1999
Y. Freund, and R. E. Schapire, 'A decision-theoretic generalization of on-line learning and an application to boosting,' Journal of Computer and System Sciences, 55 (1), 119-139, 1997

상세보기
A. Benyassine, E. Shlomot, and H.-Y. Suo 'ITU-T Recommendation G.729 Annex B: A silent compression scheme for use with G.729 optimized for V.70 digital simultaneous voice and data applications,' IEEE Communications Magazine, 64-73, Sept. 1997
H. G. Hirsch. and D. Pearce, 'The AURORA experimental framework for the performance evaluations of speech recognition systems under noisy conditions,' ISCA ITRW ASR2000 Automatic Speech Recognition: Challenges for the Next Millennium, Paris, France, Sept. 18-20, 2000
A. Acero, Acoustical and Environmental Robustness in Automatic Speech Recognition, Kluwer Academic Publishers, Boston, 1993

내보내기 구분	파일저장 인쇄 메일전송
구성항목	기본정보 상세정보 관리번호, 논문명, 저널/프로시딩명, 저자 , 발행년, 권, 호, 시작페이지, 끝페이지, 발행기관 관리번호, 논문명, 대등논문명, 저자 , 저널/프로시딩명, 발행기관, 발행년, 발행언어, 권, 호, 시작페이지, 끝페이지, ISBN, ISSN, 주제분야, 키워드, 초록(한글), 초록(영문), 저자(소속기관)
저장형식	Text(ASCII format) Excel format RefWorks Direct Export RIS format (for Reference Manager, ProCite, EndNote), Scholar's Aids, Mendeley
메일정보	받는사람 (필수) @ 보내는사람 (선택) @ 제목 내용 KISTI 검색결과 이메일 서비스
안내	총 건의 자료가 검색되었습니다. 다운받으실 자료의 인덱스를 입력하세요. (1-10,000) 검색결과의 순서대로 최대 10,000건 까지 다운로드가 가능합니다. 데이타가 많을 경우 속도가 느려질 수 있습니다.(최대 2~3분 소요) 다운로드 파일은 UTF-8 형태로 저장됩니다. 파일의 내용이 제대로 보이지 않을실 때는 웹브라우저 상단의 보기 -> 인코딩 -> 자동선택 여부를 확인하십시오. ~ Text(ASCII format) Excel format

연합인증

A Simple Speech/Non-speech Classifier Using Adaptive Boosting 원문보기

Abstract ▼ AI-Helper

주제어

AI 본문요약
AI-Helper

문제 정의

제안 방법

이론/모형

성능/효과

후속연구

참고문헌 (13)

이 논문을 인용한 문헌

저자의 다른 논문 :

관련 콘텐츠

원문 보기

이 논문과 함께 이용한 콘텐츠

AI-Helper ※ AI-Helper는 오픈소스 모델을 사용합니다.

선택된 텍스트

연합인증

A Simple Speech/Non-speech Classifier Using Adaptive Boosting 원문보기

Abstract ▼ AI-Helper

주제어

AI 본문요약 엑셀 다운로드 AI-Helper

문제 정의

제안 방법

이론/모형

성능/효과

후속연구

참고문헌 (13)

이 논문을 인용한 문헌

저자의 다른 논문 :

권오욱 (41) Lee, Te-Won (1)

관련 콘텐츠

원문 보기

이 논문과 함께 이용한 콘텐츠

AI-Helper ※ AI-Helper는 오픈소스 모델을 사용합니다.

선택된 텍스트

AI 본문요약
AI-Helper