Universal speech models for speaker independent single channel source separation

Supervised and semi-supervised source separation algorithms based on non-negative matrix factorization have been shown to be quite effective. However, they require isolated train-ing examples of one or more sources, which is often difficult to obtain. This limits the practical applicability of these al-gorithms. We examine the problem of efficiently utilizing general training data in the absence of specific training ex-amples. Specifically, we propose a method to learn a univer-sal speech model from a general corpus of speech and show how to use this model to separate speech from other sound sources. This model is used in lieu of a speech model trained on speaker-dependent training examples, and thus circum-vents the aforementioned problem. Our experimental results show that our method achieves nearly the same performance as when speaker-dependent training examples are used. Fur-thermore, we show that our method improves performance when training data of the non-speech source is available. Index Terms — source separation, non-negative matrix factorization, block sparsity 1