Noise-suppression algorithms for improved speech intelligibility by normal-hearing and cochlear implant listeners.
Philipos C. Loizou · The Journal of the Acoustical Society of America · 2010
Much research in the past few decades focused on the development of noise reduction algorithms that can suppress background noise. While these single-microphone based algorithms have been proven to improve the subjective speech quality, they have not been effective in improving speech intelligibility. This is partly due to the fact that most noise-suppression algorithms introduce speech distortion and partly because most algorithms are not optimized to operate in a particular noisy environment. Furthermore, none of the existing noise-reduction algorithms was designed to optimize a metric that correlates highly with intelligibility. This talk will present intelligibility data collected with normal-hearing and cochlear implant listeners who were presented with noisy speech processed by environment-optimized algorithms. It will also present algorithms that were designed using metrics that correlate highly with speech intelligibility. The data from these studies suggest that it is possible to develop noise reduction algorithms that improve speech intelligibility provided some constraints are imposed on the design of the suppression function and/or the intended listening environment. Research supported by NIDCD/NIH.]