Abstract | ||
---|---|---|
Formulating speech separation as a binary classification problem has been shown to be effective. While good separation performance is achieved in matched test conditions using kernel support vector machines (SVMs), separation in unmatched conditions involving new speakers and environments remains a big challenge. A simple yet effective method to cope with the mismatch is to include many different acoustic conditions into the training set. However, large-scale training is almost intractable for kernel machines due to computational complexity. To enable training on relatively large datasets, we propose to learn more linearly separable and discriminative features from raw acoustic features and train linear SVMs, which are much easier and faster to train than kernel SVMs. For feature learning, we employ standard pre-trained deep neural networks (DNNs). The proposed DNN-SVM system is trained on a variety of acoustic conditions within a reasonable amount of time. Experiments on various test mixtures demonstrate good generalization to unseen speakers and background noises. |
Year | DOI | Venue |
---|---|---|
2013 | 10.1109/TASL.2013.2250961 | IEEE Transactions on Audio, Speech, and Language Processing |
Keywords | Field | DocType |
speech processing,kernel support vector machines,relatively large datasets,monaural speech separation,binary classification problem,feature learning,acoustic conditions,dnn system,scaling up classification-based speech separation,computational complexity,linear svm,unseen speakers,deep belief networks,background noises,standard pretrained deep neural networks,very large databases,training set,discriminative features,test mixtures,support vector machines,neural nets,computational auditory scene analysis (casa),neural networks,noise,speech,acoustics,feature extraction,kernel | Kernel (linear algebra),Speech processing,Linear separability,Pattern recognition,Binary classification,Computer science,Support vector machine,Speech recognition,Artificial intelligence,Artificial neural network,Feature learning,Computational complexity theory | Journal |
Volume | Issue | ISSN |
21 | 7 | 1558-7916 |
Citations | PageRank | References |
95 | 3.90 | 27 |
Authors | ||
2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Yu-Xuan Wang | 1 | 650 | 32.68 |
DeLiang Wang | 2 | 3933 | 362.87 |