Keystroke Resilience in Laptop Acoustics Using Wiener Beamformer and Spectral Subtraction
Independent thesis Advanced level (degree of Master (Two Years))Student thesis
Nowadays, Laptops and Personal computers are increasingly being used to capture audio in various communication scenarios such as video conferencing, recordings of meetings, VOIP communications and lectures for archival purposes and audio/video instant messaging. Many of these scenarios often face unique problem of additive noise that of the user simultaneously typing on the keyboard. For example, to take notes while recording a meeting using the laptops local microphone array. The quality of speech signals captured by microphones is severely distorted by Keystroke noise which degrades the quality of speech. These degradation's lower the intelligibility of speech signal, So that listener’s ability to understand the message suffers. Thus there is a need to suppress the keystroke noise in recorded speech signals that results in significant perceptual improvement. In this thesis, the enhancement of speech signal by the suppression of keystroke noise is done through a unique combination of two algorithms i.e. Wiener Beamformer and Spectral Subtraction by Geometrical Approach. Wiener Beamformer is implemented based on subband approach. Both the algorithms individually and unique combination of both are implemented and performance is compared relative to each other by considering different performance measures i.e. Signal to Noise ratio (SNR) Improvement and PESQ. All the systems are tested with various positions of noise and speech source with reference to the microphone array and distance from speech source to microphone array is also varied. Analysing the results, the unique combination of both the algorithms proved to be efficient in terms of SNR improvement when compared to individual systems. The Spectral Subtraction algorithm does not improve the speech intelligibility effectively whereas Wiener Beamformer as an individual system works perfectly with keystroke noise suppression in speech signal and have relatively high PESQ values when compared to the values of combined system. Finally, the Spectral Subtraction reduces PESQ significantly at low SNR, the Spectral Subtraction adds few dB in improvement when combined with Weiner Beamformer, but at very high cost of decrement of PESQ.
Place, publisher, year, edition, pages
2012. , 69 p.
Wiener Beamformer, Spectral Subtraction, Filter Banks, FD Filters.
IdentifiersURN: urn:nbn:se:bth-5784Local ID: oai:bth.se:arkivex2671B1A8F7039E12C12579F900115DFFOAI: oai:DiVA.org:bth-5784DiVA: diva2:833186
Grbić, Dr. Nedelko