BI-MAGNITUDE PROCESSING FRAMEWORK FOR NONLINEAR ACOUSTIC ECHO CANCELLATION ON ANDROID DEVICES

This paper presents a new paradigm for acoustic echo control on mobile Android devices. The echo path on these devices has nonlinearities including not only the results of overdriven power amplifiers and miniaturized loudspeakers, but also those caused by hardware audio dynamic range compressor (ADRC). While the former form of nonlinearities was widely investigated in past research, the latter has not yet been taken into account. The ADRC adds extra gains to the echo path and makes it become a fast time-varying system. This presents a great challenge to traditional (both linear and nonlinear) echo cancellation systems. Here we propose a novel bi-magnitude processing framework, which is based on a two-state model for the echo path. The algorithm can deal with the ADRC problem and offers robust control for identification of input nonlinearities. The performance of the proposed approach is evaluated on recordings made in an anechoic chamber using real Android devices.

• Machine Perception

• Speech Processing