A Multi-filter Approach to Acoustic Echo Cancellation for Teleconferencing

We have designed and tested an acoustic echo cancellation system for speech teleconferencing. Our algorithm is based on a least-mean-square (LMS) frequency domain adaptive filter (FDAF) and uses a novel filter-update technique using many (at least 3) simultaneously running filters. We find the new multi-filter to converge faster than similar LMS FDAF’s for echo cancellation, and find it to be especially robust during double-talk conditions.