Design of a combined air and bone-conductive microphone headset

Project Type: Master/Diploma Thesis
Student: Domes Christoph

 

 In real-life environments speech signals can easily be degraded by ambient interferences, e.g. other competing speakers, speech babble, or ambient noise. In automatic speech recognition, for example, these interferences can reduce the number of correctly recognized words dramatically. In general, speech enhancement and voice activity detection algorithms are used to process the input speech prior to recognition. However, the detection of words spoken by the target speaker is still an open issue. In this work, besides the conventional air microphone an additional bone-conductive microphone is investigated to tackle the target speaker detection problem. The bone-microphone is directly placed on the skull of the head to capture structure-borne signals. This structure-borne signal is a time-variant low-pass filtered version of the speech signal but is insensitive to further ambient noise. This thesis deals with the design of a combined air and bone-conductive microphone headset. Further, an USB audio interface to record the two microphone speech signals with the headset on the computer was developed. This interface consists of two separate microphone channels with a preamplifier, a headphone output and different power supplies for the microphones. The hardware was used to record a speech database. Speech recordings with different speakers and different interferences were made and analysed. The additional signal of the bone-conductive microphone is utilized to make Voice Activity Detection (VAD) more robust.