[논문]가중특징 Mahalanobis거리를 이용한 마이크 어레이 음석인식의 성능향상

정현열

가중특징 Mahalanobis거리를 이용한 마이크 어레이 음석인식의 성능향상
Performance Improvement of Microphone Array Speech Recognition Using Features Weighted Mahalanobis Distance 원문보기

The journal of the Acoustical Society of Korea, v.29 no.1E, 2010년, pp.45 - 53

(Department of Information and Communication, Yeungnam University) , 정현열 (Department of Information and Communication, Yeungnam University)

Abstract ▼ AI-Helper

In this paper, we present the use of the Features Weighted Mahalanobis Distance (FWMD) in improving the performance of Likelihood Maximizing Beamforming (Limabeam) algorithm in speech recognition for microphone array. The proposed approach is based on the replacement of the traditional distance measure in a Gaussian classifier with adding weight for different features in the Mahalanobis distance according to their distances after the variance normalization. By using Features Weighted Mahalanobis Distance for Limabeam algorithm (FWMD-Limabeam), we obtained correct word recognition rate of 90.26% for calibrate Limabeam and 87.23% for unsupervised Limabeam, resulting in a higher rate of 3% and 6% respectively than those produced by the original Limabearn. By implementing a HM-Net speech recognition strategy alternatively, we could save memory and reduce computation complexity.

주제어

AI 본문요약
AI-Helper

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

문제 정의

In this paper, wepresented the results of our investigation on the recognition performance of Limabeam applied for Korean speech.

가설 설정

1. Reducing the required memory for storing HMM model: Usually, the size of LogMelSpec HMMs is about 2 times larger than that of MFCC HMMs.
4. Determine optimal state sequence through the calibrated transcription, array outp나f signal, and HMM from speech recognizer.
6. Go to step 3 if 匕(g) has not converged.

제안 방법

Seltzer overcame this problem by proposing Limabeam [5]. In this algorithm, Seltzer broke the pipeline structure of classical microphone array processing techniques by using the information from the speech recognition system itself to find the array parameters that improve the speech recognition performance. Figure 1 shows the structure of this approach.
Optim기 state sequence was estimated on cepstral domain and the LogMelSpec HMM Model was calculate directly from MFCC HMM mod애 through IDCT.
The system was trained, using 39-dimensional feature vectors consisting of 13 Mel Frequency Cepstral Coefficients (MFCC) parameters, along with their delta and delta-delta parameters.

대상 데이터

A total of 596 utterances produced by six speakers recorded in above environment were used for testing.
The data were recorded in a room where the noise came from sources which included computer fans, air-conditioner, h니man voice, footsteps and slams outside.

이론/모형

Flowchart of proposed scheme to implement Limabeam algorithm.

성능/효과

2. Reducing complexity'- Generating a LogMelSpec parallel set of HMMs requires converting speechsignals to log-mel spectral features process and employing STAR algorithm to generate parallel models.
Also, our results indicated that recognition performance could be improved when the filter length was increased.
In the tests, calibrate Limabeam gave better recognition performance than unsupervised Limabeam, However, as we know, calibrate Limabeam need to have utterances to adapt the FIR parameters to every recognition environment but unsupervised Limabeam does not need them.
The results showed that recognition performance of FWMD-Limabeam was better than distance measure in original Limabeam algorithm because the drawback of the Mahalanobis distance in original Limabeam was the equal adding up of the variance normalized squared distances of the features.
When we appli언d our proposed FWMD to both calibrated and unsupervised Limabeam, we obtained increased recognition accuracy, with a correct word recognition rate of 90.26 % for FWMD—calibrate Limabeam and of 87.23 % for FWNlD-unsupervised Limabeam respectively, resulting in approximately an increase rate of 3 % and 6 % higher than that of conventional Limabeam.

참고문헌 (10)

L. J. Griffiths and C. W. Jim, "An alternative approach to linearly constrained adaptive beamforming," IEEE Transaction on Antennas and Propagation, vol, AP-30, no. 1, pp. 27-34, January 1982.
R. Zelinski, "A microphone array with adaptive post-filtering for noise reduction in reverberant room." ICASSP-88, vol. 5, pp. 2578-2581, 1988.
I. A. McCowan and H. Bourlard, "Microphone array post-filter for diffuse noise filed," ICASSP 2002, vol. 1, pp. 905-908, 2002.
S. Leukimmiatis, "An Optimum Microphone Array Post-Filter for Speech Application," ICSLP-INTERSPEECH, pp. 2142-2145, Sep. 2006.
M. Seltzer, "Microphone array processing for robust speech recognition,' Doctoral dissertation, Carnegie Mellon University, Pittsburgh, PA, 2003.
M. Wolfel and H. K. Ekenel, "Feature weighted Mahalanobis distance: improved robustness for Gaussian classifiers," 13th European Signal Processing conference EUSIPCO2006, Antalya, Turkey, Sep. 2005.
C. H. Knapp and C. Carter, "The generalized correlation method for estimation of time delay," IEEE Transactions on Information Theory, vol. 13, no. 2, pp. 260-269, April 1967.

상세보기
A. J. Viterbi, "Error bounds for convolution codes and an asymptotically optimum decoding algorithm," IEEE Transactions on Information Theory, Vol. 13, inssue 2. pp.260-267, April 1967.
M. Suzuki, S. Makino, A. Ito, H. Aso and H. Shimodaira, "A new HMnet construction algorithm requiring no contextual factors," IEICE Trans, On Information System, Vol. E78-D, No 6, pp. 662-669, 1995.
N. T. Hieu, "Robust Speech Recognition using Microphone Array in Adverse Environment," M,S Thesis, Yeungnam University, 2006.

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

가중특징 Mahalanobis거리를 이용한 마이크 어레이 음석인식의 성능향상
Performance Improvement of Microphone Array Speech Recognition Using Features Weighted Mahalanobis Distance 원문보기

Abstract ▼ AI-Helper

주제어

AI 본문요약
AI-Helper

문제 정의

가설 설정

제안 방법

대상 데이터

이론/모형

성능/효과

참고문헌 (10)

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

가중특징 Mahalanobis거리를 이용한 마이크 어레이 음석인식의 성능향상 Performance Improvement of Microphone Array Speech Recognition Using Features Weighted Mahalanobis Distance 원문보기

Abstract ▼ AI-Helper

주제어

AI 본문요약 엑셀 다운로드 AI-Helper

문제 정의

가설 설정

제안 방법

대상 데이터

이론/모형

성능/효과

참고문헌 (10)

이 논문을 인용한 문헌

저자의 다른 논문 :

Nguyen, Dinh Cuong (2) 정현열 (45)

관련 콘텐츠

원문 보기

이 논문과 함께 이용한 콘텐츠

AI-Helper ※ AI-Helper는 오픈소스 모델을 사용합니다.

선택된 텍스트

가중특징 Mahalanobis거리를 이용한 마이크 어레이 음석인식의 성능향상
Performance Improvement of Microphone Array Speech Recognition Using Features Weighted Mahalanobis Distance 원문보기

AI 본문요약
AI-Helper