An approach to refine and adapt an existing music sound source separation algorithm to speech enhancement is presented. The existing algorithm has the capability to extract music sources from stereo recordings using the position of the sources in the stereo field. Described in this paper is the ability of a modified Azimuth Discrimination and Resynthesis algorithm (m-ADRess) to enhance speech in the presence of noise using a two-microphone array. Also proposed is a novel extension to the algorithm, which enables further noise removal from speech based on elevation angle of arrival. Objective measures and an informal listening test of processed speech show the suitability of m-ADRess for cleaning noisy speech mixtures in an anechoic environment.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.