COMPARISON OF REFERENCE MICROPHONE SELECTION ALGORITHMS FOR DISTRIBUTED MICROPHONE ARRAY BASED SPEECH ENHANCEMENT IN MEETING RECOGNITION SCENARIOS

This paper addresses a front-end system for speech recognition of meeting conversations that are recorded with distributed microphones such as smartphones. When using distributed microphones, one of the microphones may be closer to the speaker than the others and thus provide high speech recognition accuracy due to a high signal-to-noise ratio and low reverberation. It is important to select such a microphone as a reference microphone channel in widely studied speech enhancement approaches, which estimate source images at a reference microphone. However, the reference microphone selection is still an open problem, especially for a distributed microphone array, where the sensitivity may differ among the microphones. In this paper, we discuss several approaches to select a reference microphone for multi-channel speech enhancement, such as independent vector analysis (IVA), and compare the performance of these approaches in terms of speech recognition accuracy.