http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Automatic Music Summarization Using Similarity Measure Based on Multi-Level Vector Quantization
김성탁,김상호,김회린,Kim, Sung-Tak,Kim, Sang-Ho,Kim, Hoi-Rin The Acoustical Society of Korea 2007 韓國音響學會誌 Vol.26 No.e2
Music summarization refers to a technique which automatically extracts the most important and representative segments in music content. In this paper, we propose and evaluate a technique which provides the repeated part in music content as music summary. For extracting a repeated segment in music content, the proposed algorithm uses the weighted sum of similarity measures based on multi-level vector quantization for fixed-length summary or optimal-length summary. For similarity measures, count-based similarity measure and distance-based similarity measure are proposed. The number of the same codeword and the Mahalanobis distance of features which have same codeword at the same position in segments are used for count-based and distance-based similarity measure, respectively. Fixed-length music summary is evaluated by measuring the overlapping ratio between hand-made repeated parts and automatically generated ones. Optimal-length music summary is evaluated by calculating how much automatically generated music summary includes repeated parts of the music content. From experiments we observed that optimal-length summary could capture the repeated parts in music content more effectively in terms of summary length than fixed-length summary.
김성탁,지미경,김회린,Kim, Sung-Tak,Ji, Mi-Kyong,Kim, Hoi-Rin 한국음향학회 2009 韓國音響學會誌 Vol.28 No.3
Speaker verification determines whether the claimed speaker is accepted based on the score of the test utterance. In recent years, methods based on Gaussian mixture models and universal background model have been the dominant approaches for text-independent speaker verification. These speaker verification systems based on these methods provide very good performance under laboratory conditions. However, in real situations, the performance of speaker verification system is degraded dramatically. For overcoming this performance degradation, the feature recombination method was proposed, but this method had a drawback that whole sub-band feature vectors are used to compute the likelihood scores. To deal with this drawback, a modified feature recombination method which can use each sub-band likelihood score independently was proposed in our previous research. In this paper, we propose a sub-band weighting method based on sub-band signal-to-noise ratio which is combined with previously proposed modified feature recombination. This proposed method reduces errors by 28% compared with the conventional feature recombination method. 화자검증은 발성화자가 제시화자 (claimed speaker)인지 아닌지를 구별하는 것이다. 기존의 화자검증 시스템인 GMM-UBM 방식의 화자검증 시스템은 무잡음 환경에서는 높은 검증성능을 보이지만, 잡음환경에서는 성능이 급격히 떨어지는 단점이 있다. 이런 단점을 극복하기 위해 멀티밴드를 이용한 방법인 특징벡터 재결합방법이 제안되었지만, 특징벡터 재결합방법은 전체 서브밴드 특징벡터들을 사용하여 유사도를 계산하는 단점이 있다. 이런 단점을 극복하기 위해 기 발표된 이전 논문에서 각 서브밴드 유사도를 독립적으로 계산하는 변형된 특징벡터 재결합방법을 제안하였고, 본 논문에서는 변형된 특징벡터 재결합방법과 각 서브밴드들의 신뢰도를 나타내는 신호 대 잡음비를 이용한 가중치를 이용하여 잡음환경에서 기존의 특징벡터 재결합방법에 비해 에러를 28% 감소시켰다.
Optimal Decision Tree를 이용한 Unseen Model 추정방법
김성탁,김회린,Kim Sungtak,Kim Hoi-Rin 대한음성학회 2003 말소리 Vol.45 No.-
Decision tree-based state tying has been proposed in recent years as the most popular approach for clustering the states of context-dependent hidden Markov model-based speech recognition. The aims of state tying is to reduce the number of free parameters and predict state probability distributions of unseen models. But, when doing state tying, the size of a decision tree is very important for word independent recognition. In this paper, we try to construct optimized decision tree based on the average of feature vectors in state pool and the number of seen modes. We observed that the proposed optimal decision tree is effective in predicting the state probability distribution of unseen models.