Multi Channel Speech Separation Essay

1. INTRODUCTION This write up is a synopsis of the work attempted to develop the techniques for Multi channel speech separation, to separate more than one speech signal from a single mixture captured in both stationary and non-stationary noisy environment. Single channel speech separation separates the single speech signal from the noisy environment. One of the most powerful techniques for speech enhancement in single channel and co-channel is an adaptive noise cancellation. It adapts the process of filtering the noise using filters. But this technique is not applicable when the interference of speech signal contains another speech signal. Co-channel speech separation aims to separate the two clean speech signals transmitted on the same channel

But, if it is a speech signal, it is not possible to do so. Hence the properties of speech in the mixture are to be studied and are used in speech separation. This separation of speech signals from the speech mixture can be done by either supervised technique or using unsupervised technique. For all the supervised learning algorithms, clean utterances should be available apriori for the system to construct speaker dependent models. One of the main drawbacks of supervised learning is the mismatch between training and testing signals. The Unsupervised learning technique needs no prior information about the speech or speaker and separates the speech using the information available from the speech mixture and is efficient in speech separation. Existing approaches for multi channel speech separation were mainly based on multiple sources captured using multiple microphones and using supervised approach.
2. OBJECTIVES To develop a multi channel speech separation using unsupervised technique in clean and noisy environment with reduced time complexity to apply in real world applications for hearing impaired persons. And to evaluate the efficiency of these new approaches in speech separation techniques.
3. LITERATURE