Maximum Likelihood Noise Cancellation Using the EM (Estimate-Maximize) Algorithm

摘要

Single-microphone speech enhancement systems have typically shown limited performance. Two-microphone systems based on a least-squares error criterion have shown better results in some contexts; however, sometimes the desired (speech) signal is cancelled together with the noise. In this paper we suggest a new approach to the two-microphone speech enhancement problem. Specifically, we formulate a maximum likelihood (ML) problem for estimating the parameters needed for cancelling the noise, and then, we solve this ML problem via the iterative EM (Estimate-Maximize) technique. The resulting procedure shows encouraging results that improve upon the 'classical' least-squares approach. Reprints. (JHD )