Attention-long short term memory 기반의 화자 임베딩과 I-vector를 결합한 원거리 및 잡음 환경에서의 화자 검증 알고리즘 = Speaker verification system combining attention-long short term memory based speaker embedding and I-vector in far-field and noisy environments

국문 초록 (Abstract)

문장 종속 짧은 발화에서 문장 독립 긴 발화까지 다양한 환경에서 I-vector 특징에 기반을 둔 많은 연구가 수행되었다. 본 논문에서는 원거리 잡음 환경에서 녹음한 데이터에서 Probabilistic Linear Discriminant Analysis (PLDA)를 적용한 I-vector와 주의 집중 기법을 접목한 Long Short Term Memory(LSTM) 기반의 화자 임베딩을 추출 하여 결합한 화자 검증 알고리즘을 소개한다. LSTM 모델의 Equal Error Rate(EER)이 15.52 %, Attention-LSTM 모델 이 8.46 %로 7.06 % 성능이 향상되었다. 이로써 본 논문에서 제안한 기법이 임베딩을 휴리스틱 하게 정의하여 사용하 는 기존 추출방법의 문제점을 해결할 수 있는 것을 확인하였다. PLDA를 적용한 I-vector의 EER이 6.18 %로 결합 전 가장 좋은 성능을 보였다. Attention-LSTM 기반 임베딩과 결합하였을 때 EER이 2.57 %로 기존보다 3.61 % 감소하 여 상대적으로 58.41 % 성능이 향상되었다.

다국어 초록 (Multilingual Abstract)

Many studies based on I-vector have been conducted in a variety of environments, from text- dependent short-utterance to text-independent long-utterance. In this paper, we propose a speaker verification system employing a combination of I-vector with Probabilistic Linear Discriminant Analysis (PLDA) and speaker embedding of Long Short Term Memory (LSTM) with attention mechanism in far-field and noisy environments. The LSTM model’s Equal Error Rate (EER) is 15.52 % and the Attention-LSTM model is 8.46 %, improving by 7.06 %. We show that the proposed method solves the problem of the existing extraction process which defines embedding as a heuristic. The EER of the I-vector/PLDA without combining is 6.18 % that shows the best performance. And combined with attention-LSTM based embedding is 2.57 % that is 3.61 % less than the baseline system, and which improves performance by 58.41 %.

참고문헌 (Reference)

• 1 D. A. Reynolds, "Speaker verification using adapted gaussian mixture models" 10 : 19-41, 2000
• 2 Q. Wang, "Speaker diarization with lstm" 5239-5243, 2018
• 3 S. J. D. Prince, "Probabilistic linear discriminant analysis for inferences about identity" 1-8, 2007
• 4 D. Bahdanau, "Neural machine translation by jointly learning to align and translate"
• 5 P. Kenny, "Joint factor analysis versus eigenchannes in speaker recognition" 15 : 2072-2084, 2007
• 6 B. Fauve, "Improving the performance of text-independent short duration SVMand GMM based speaker verification" 18-, 2008
• 7 L. Wan, "Generalized end-to-end loss for speaker verification"
• 8 N. Dehak, "Front-end factor analysis for speaker verification" 19 : 788-798, 2011
• 9 G. Heigold, "End-toend text-dependent speaker verification" 5115-5119, 2016
• 10 E. Variani, "Deep neural networks for small footprint text-dependent speaker verification" 4080-4084, 2014
• 11 D. Snyder, "Deep neural network embeddings for text-independent speaker verification" 999-1003, 2017
• 12 Y. Liu, "Deep feature for text-dependent speaker verification" 73 : 1-13, 2015
• 13 F. Chowdhury, "Attention-based models for text-dependent speaker verification"
• 14 V. Peddinti, "A time delay neural network architecture for efficient modeling of long temporal contexts" 3214-3218, 2015