Speech Localization at Low Bitrates in Wireless Acoustics Sensor Networks

The use of speech source localization (SSL) and its applications offer great possibilities for the design speaker local positioning systems with wireless acoustic sensor networks (WASNs). Recent works have shown that data-driven front-ends can outperform traditional algorithms SSL when trained to work in specific domains, depending on factors like reverberation noise levels. However, such models consider directly from raw observations, without consideration transmission losses WASNs. In contrast, sensors reside separate real-life devices, we need quantize, encode transmit data, decreasing performance localization, especially bitrate is low. this work, investigate effect low a Direction Arrival (DoA) estimator. We analyze deep neural network (DNN) based framework as function audio encoding compressed signals by employing recent communication codecs including PyAWNeS, Opus, EVS, Lyra. Experimental results show training DNN input encoded PyAWNeS codec at 16.4 kB/s improve accuracy significantly, up 50% degradation almost all be recovered. Our further best model one two channels higher than 32 kB/s, it optimal data second channel. lower bitrate, preferable similarly channels. More importantly, practical applications, more generalized randomly selected each channel, shows large gain least PyAWNeS.