확률적 매칭 방법을 이용한 강인한 감정 음성 인식
Robust Emotional Speech Recognition using Stochastic Matching Method
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
다양한 감정을 지닌 음성은 음성 인식 시스템의 성능을 저하시킨다. 본 논문에서는 적은 개수의 감정 음성을 사용하여 음성 인식시스템을 감정 변화에 강인하게 만들기 위하여 확률적 매칭 방법을 사용한 감정 적응 기법으로 감정 모델을 생성하는 방법을 제안하였다. 기존 방법은 많은 개수의 감정 학습 데이터가 필요하였으나 본 연구에서는 감정 적응을 위하여 적은 개수의 학습 감정 음성과감정 적응 방식을 사용하여 감정이 없는 음성 모델로부터 감정 음성 모델이 생성되었다. 학습을 위하여 화자로부터 충분한 감정 음성을 얻는 것이 어렵기 때문에 적은 개수의 감정 음성을 사용하여 학습할 수 있다는 것은 실제 상황에서 매우 실용적이다. 확률적매칭 방법은 입력 파라메터에 포함된 감정 변화를 제거하는 파라메터 영역 제거 방법과 학습 과정에서 적은 수의 감정 데이터를 사용하여 평상 감정의 음성 모델을 감정 음성 모델로 감정 적응하는 두 가지 방법으로 적용되었다. 실험 결과에서 두 가지 방법 모두가 감정에 의해 발생하는 오류를 감소시켰으며 두 가지 방법을 함께 사용하면 오차를 더욱 줄일 수 있었다.
Abstract
Speech with various emotions degrades the performance of speech recognition systems. This paper proposes amethod to create an emotion model using a emotional adaptation technique with a stochastic matching method tomake the speech recognition system robust to emotional changes using a small number of emotional speechsamples. Traditional methods required a large number of emotional training data, but this study generates anemotional speech model from a neutral speech model using a small number of emotional training speech samplesand emotional adaptation techniques. Since it is difficult to obtain sufficient emotional speech from speakers fortraining, the ability to train with a small number of emotional speech samples is very practical in real situations. The stochastic matching method was applied in two ways: a parameter domain removal method to removeemotional changes included in the input parameters and an emotional adaptation method to adapt from a neutralemotion speech model to an emotional speech model using a small amount of emotional training data during thelearning process. Experimental results showed that both methods reduced errors caused by emotions, and whenboth methods were used together, the errors were further reduced.
참고문헌 (17)
- Benzeghiba, M, "Automatic speech recognition and speech variability : a review", Speech Comm, 49(10), pp.763~786 (2007)[DOI]
- Cowie, R, "Emotion recognition in human-computer interaction", IEEE Signal Proc. Mag, 18(1), pp.32~80 (2001)[DOI]
- Xiong, X, "Robust speech features and acoustic models for speech recognition", Ph. D, pp.2009
- Sun, Y, "Acoustic feature optimization for emotion affected speech recognition", Proc. International Conference on Information Engineering and Computer Science, pp.1~4 (2009)
- Lee G., "Robust speech parameters for the emotional speech recognition", Journal of the Korea Institute of Intelligent Systems, 22, pp.681~686 (2012)[DOI]
- Sheikhan, M, "Using DTW neural–based MFCC warping to improve emotional speech recognitio", Neural Comput. Appl, 21(7), pp.1765~1773 (2012)[DOI]
- Bashirpour M, "Robust emotional speech recognition based on binaural model and emotional auditory mask in noisy environments", EURASIP Journal on Audio, Speech, and Music Processing, 09(1), pp.1~13 (2018)[DOI]
- Pan, Y, "Emotion-detecting based model selection for emotional speech recognition", Proc. Multiconference on Computational Engineering in Systems Applications, pp.2169~2172 (2006)
- Ijima, Y, "Emotional speech recognition based on style estimation and adaptation with multiple-regression HMM", Proc. ICASSP 2009, pp.4157~4160 (2009)
- Ijima, Y, "A rapid model adaptation technique for emotional speech recognition with style estimation based on multiple-regression HMM", IEICE Trans. Inf. & Syst, E93-D(1), pp.107~115 (2010)[DOI]
- W. Kim, "A Training Method for Emotionally Robust Speech Recognition using Frequency Warping", Journal of the Korea Institute of Intelligent Systems, 20, pp.530~535 (2010)[DOI]
- J. Koehler, "Integrating RASTA-PLP into Speech Recognition", Proc. ICASSP, pp.421~424 (1994)
- M. G. Rahim, "Signal Bias Removal by Maximum Likelihood Estimation for Robust Telephone Speech Recognition", IEEE Trans. Speech & Audio Processing, 4(1), pp.19~30 (1996)[DOI]
- Sankar, A., "A Maximum-Likelihood Approach to Stochastic Matching for Robust Speech Recognition", IEEE Trans. on Speech and Audio Processing, 4, pp.190~202 (1996)[DOI]
- C. J. Leggetter, "Maximum likelihood linear regression for speaker adaptation of continuous density hidden Markov models", Comput. Speech Lang, 9(2), pp.171~185 (1995)[DOI]
- C. H. Lee, "A study on speaker adaptation of the parameters of continuous density hidden Markov models", IEEE Trans. on Signal Processing, 39(4), pp.806~814 (1991)
- B. S. Kang, "Context Independent Emotion Recognition System using Speech Signal", Master, pp.2000