摘要 |
A speech detector (200, fig. 2) receives a near-end microphone signal (201) containing both near-end speech and a contribution h from a far-end signal (208) acoustically coupled to the microphone via output speaker (202), and estimates the pure nearend speech signal (205) by transforming the microphone 302 and farend 301 signals into the frequency domain 304, 305, measuring the gain between the microphone signal and the nearend signal 307, and classifying speech 311 in dependence on the variance of the gain and nearend signal. The output 313 may then be fed to an Adaptive Echo Estimation Filter (AEEF, 203) to cancel the echo. |