Acoustic Fan Noise Cancellation in a Laptop Microphone System

Abstract: Speech communication involving audio conferencing, video conferencing, teleconferencing via laptops became greatly influenced in office environments i.e. between employer and employee, and also influenced in personal life meetings between friends or in-between parents and children. These meeting conversations will mostly disturbs by annoying noise, i.e. fan noise which is produced by laptop cooling fan, which suffers at the both ends of communication due to this noise. With this noise effect the intelligibility of original speech is degraded between the conversations of meetings. So there is need of enhancing the speech from noisy speech environment in the communication. Thus speech enhancement is emerging technology in the communication and signal processing filed. So this thesis focuses on attenuating the noise produced by laptop cooling fan, with use of different speech enhancement algorithms. In this thesis we implement a multichannel Microphone Array (MA) of linearly arranged two microphones with different speech enhancement algorithms in spatial frequency domain. As the implementation involves frequency domain, we design a filter bank which is back bone structure of thesis, which is used to transform the signals received by microphone array into subband sequence, this subbands are processed through enhancement algorithms to attenuate noise, and then finally used to reconstruct the estimated original speech signal in time domain. The speech enhancement algorithms, involve beamforming technique i.e. Wiener BeamForming (WBF), and Spectral Subtraction (SS). Here we utilize a Direction Of Arrival (DOA) technique, to localize the speech source based on Time Difference Of Arrival (TDOA) in frequency domain only. Here we implements different systems involving individual WBF and SS algorithms, and also hybrid combination of algorithms WBF and SS, to suppress the fan noise. These systems were implemented at different positions of speech and noise sources. These systems were implemented and evaluated using simulation tool Matlab. The objective quality measures used to validate the systems are Signal to Noise Ratio Improvement (SNRI) and Perceptual Evaluation Speech Quality (PESQ) measure. The systems were validated with a pure speech combination of male and female sampled at 16 KHz, and fan noise recorded in the real time of anechoic environment. The systems are simulated at different SNR ratios of 0dB, 5dB, 10dB, 15dB, 20dB. The simulation result shows that hybrid system proves to be efficient in reduction of noise at higher SNR ratios, with compromise of speech quality. Whereas the individual beamformer system proves to be very highly efficient in reducing the noise while maintaining high quality in speech at both high and low SNR’s, whereas the spectral subtraction individual alone reduces a noise sufficiently only but its speech quality is very poor in performances. Also at different positions the systems were simulated, at the position where DOA of noise is 90 then all systems works highly efficient in reducing noise since SNRI are around 40dB to 50 dB while maintaining speech quality.