Collaborative voice activity detection for hearing aids

Most binaural hearing aid systems have separate processing for each ear. Collaborative processing allows the left and right hearing aids to share information and resources. In this paper, collaborative processing is used to take advantage of head shadow effects to improve Voice Activity Detection (VAD). We emphasize here that while a variety of improvements in VAD and speech enhancement are possible, contributions have meaning only if the advances fit into the framework of the limited computing resources available in current hearing aid devices. The proposed approach leads to an absolute speech hit rate improvement of up to 25% at low SNR in the ear closest to noise source. The cost is a lower noise hit rate in the other ear. It is however sufficient to ensure proper noise spectrum magnitude estimation.