Abstract

Typically, users of teleconferencing desktop applications must use headphones to eliminate the direct-path echo introduced by having a loudspeaker and a microphone in the same room. Unfortunately, many users prefer not to use headphones. Implementing echo cancelation would be one way for desktop audio conferencing systems to overcome this limitation. One technique used to remove the echo is to introduce an autoregressive filter to identify the echo path impulse response defined by the system fSpeaker + Room + Microphoneg. The signal output to the speaker is then filtered by the estimated echo path impulse response, and the result is subtracted from the degraded input speech to obtain an estimate of the input speech without echoes. In this paper, we describe the identificationprocess which leads to the least mean square (LMS) algorithms and the recursive least square (RLS) algorithms. We also present an "Echo-Cancelation Software Lab" whichwas implemented and optimized to allow real-tim...

Citations