A microphone array for echo cancellation in audioconference rooms

Abstract

This paper describes an adaptive acoustic antenna (or microphone array) designed for sound recording in audioconferencing rooms. This microphone array achieves three goals simultaneously: the sound from the useful sources (speakers) is recorded without distortion, several jammers emitting known signals can be cancelled (this is the case for the sound received from the other room and emitted by the loudspeakers), and finally, the ambient noise and the room reverberation are reduced. This last effect enforces the “presence effect” in the recordings. In the present approach, echo cancellation is seen as the problem of rejecting a jammer, and it is done by signal combination rather than by subtraction. To achieve these goals, the microphone array uses adaptive beamforming techniques. The beamformer is a compromise between data‐independent or fixed beamformers and data‐dependent or adaptive ones. This makes it robust with respect to localization errors: the beamformer uses an identified localization of the jammers but weighs this information by some knowledge of the signal‐to‐noise ratio. The localization of the main jammer (the echo coming from the loudspeaker) requres the identification of a long impulse response (the acoustic channel between the loudspeaker and each microphone). A model to identify this response in the time‐frequency domain is developed. This model involves a “double convolution” and its adaptive identification has some analogies with subband adaptive methods. The estimated parameters of this model are used to compute the optimal weights for the beamformer every 8 ms. The paper will present simulations of these techniques based upon a first prototype that consists of an array of 15 microphones on a semicircle with diameter 1 m. The performances of this adaptive microphone array are characterized in terms of echo rejection, convergence rate, and level of dereverberation.