Media.Player

Improved Echo-path change Detection for Acoustic Echo Cancellation

详细技术说明

None

*Abstract

Non-Confidential Abstract of Invention: Most teleconferencing conversations are conducted in the presence of acoustic echoes. An acoustic echo canceller (AEC) is used to remove the echo created due to the loudspeaker-microphone environment as shown in Figure 1. In an AEC the echo-path is adaptively modeled using a filter, which is then used to synthesize a replica of the echo. This synthesized replica of the echo is subtracted from the echo-corrupted microphone signal to get an echo-free signal. When the near-end talker is active or when the speech comes from both the far-end and near-end, the adaptive filter coefficients diverge from the true echo path impulse response if the adaptation is not halted. A doubletalk detector is used to stop the AEC's filter adaptation during periods of near-end speech. A Doubletalk detector should be able to detect a doubletalk condition quickly and accurately so as to freeze adaptation as soon as possible; at the same time it should be able to track any echo-path changes and should be able to distinguish the doubletalk from the echo-path variations. Increased sensitivity towards doubletalk results in declaring echo-path changes as doubletalk which adversely affects the performance of an AEC as we freeze adaptation when we really need to adapt. Thus, we need an efficient and simple echo-path change detector so as to differentiate any echopath variations from doubletalk. The proposed algorithm for echo-path change detection is computationally very efficient as we only do 9 operations to compute the decision statistic at each sample as compared to (6L+4) operations for the best known existing technique where L is the frame size (Typically greater than 51 2), also our decision statistic is normalized appropriately and meets the needs of an optimal echo-path change detector. Further, for effective doubletalk control we can avoid the two-path method for echo cancellation with the aid of the proposed echo-path change detector which results in significant memory and computational savings.

*Principal Investigation

Name: Asif Mohammad

Department:

Name: Steven Grant, Professor

Department:

其他

国家/地区

美国