Speech enhancement methods in hands-free communication with emphasis on Optimal SNIR Beamformer

Sammanfattning: A basic speech enhancement can be achieved by the suppression of background noise and reverberation from the clean speech. The point to be noted is to achieve it with a low computational complexity. The aim is to estimate signal arriving optimally from the desired direction in the presence of reverberant-noisy speech signal. Recent studies show that this can be achieved by designing various kinds of robust fixed and adaptive beamformers. A beamformer does spatial filtering in the sense that it separates two signals with overlapping frequency content originating from distinctive directions. In this contribution, robust beamformers namely Elkos beamformer, Wiener beamforming and optimal signal to noise interference ratio (SNIR) beamformer are designed and analyzed collaboratively in a group under the consideration of hearing aid constraints such as the microphone distance and different real world room dimensions. A fractionally delayed (FD) all pass Thiran filters are designed to get a maximally flat group delay. A virtual room image model is designed to achieve different dimensions of the room and their reverberant speech signals. The objective of this thesis is to design and implement an optimal SNIR beamformer in anechoic and reverberant environments with different noises, i.e. wind, white, factory and interference. It is implemented and simulated offline in MATLAB. The performance of the optimal SNIR Beamformer is evaluated by considering the objective measures such as SNRI, SD, ND, RR and PESQI under different noisy environments in anechoic and reverberated environments. These parameters are measured by assuming input SNR levels at 0dB, 5dB, 10 dB, 15 dB, 20 dB and 25 dB. In addition to this a new parameter RR is also evaluated in reverberated environment. This parameter is measured by varying the number of microphones. The reverberation power suppression is analyzed by using RR. Speech quality is analyzed based on signal to noise ratio Improvement and speech intelligibility is measured using PESQ for different noisy environments. Results show that optimal SNIR beamformer performs best compared to all other beamformers due to its inherent properties.