Beamforming and Blind Signal Separation for Far-field Voice Capture using a Microphone Array

Abstract: With the evolving technology of mobile electronics and other forms of communication methods, an increasing demand of speech intelligibility is introduced. In a conference room for example, multiple microphones are placed directly in front of the speaker in order to pick up clean speech. However, several challenges still exist. For instance the distance between speaker and microphone will not be fixed, unwanted noise from interfering sources around the room might be picked up and reverberation from the actual signal of interest might be introduced. The same problem exists whilst talking into a mobile device when the wind is blowing in the background, which might be easier to relate to. There is a need for adaptive methods which takes these parameters in to consideration in order to improve speech intelligibility. One way to solve this problem is by the use of an acoustic beam. Ideally, this will leave the source of interest untouched and suppress all unwanted noise. This can be done either by putting a beam in a predetermined direction of arrival or adaptable direction, commonly known as blind signal separation. This is achieved with help of multiple microphones working in tandem, hence the title Beamforming and Blind Signal Separation for Far-field Voice Capture using a Microphone Array. This thesis will investigate the possibility of using beamforming for far-field voice capture using a commercially available microphone array. The concepts are explained and then one direction-of-arrival and two beamforming algorithms are implemented, tested and evaluated. The direction-of-arrival algorithm is based on cross correlation and the beamforming algorithms are known as delay-and-sum and minimum variance distortionless response.