Multi-sensor sub-band adaptive noise cancellation for speech enhancement in an automobile environment

Concerning noise in automobiles, the effect of spacing sub-bands nonlinearly in a multi-microphone sub-band adaptive (MMSBA) speech enhancement system as in humans according to a published cochlear function (Greenwood, 1990) is investigated, and the MMSBA scheme is further developed by employing diverse sub-band processing (SEP) in order to allow inter-channel features within the sub-bands to influence the subsequent processing. In order to realize this, a practical metric is developed capable of real-time implementation based on the inter-channel magnitude squared coherence (MSC) relationship which is adaptively estimated during noise alone periods in intermittent speech, for automatically selecting the best SEP option. The choice of SEP options include: (i) no processing; (ii) intermittent coherent noise canceller; and (iii) continuous incoherent noise canceller. The performance of the MMSBA scheme is assessed using informal subjective listening tests for the case of real reverberant automobile data.