PROBABILISTIC APPROACH FOR SPEECH INTELLIGIBILITY IMPROVEMENT AND NOISE REDUCTION
D. Sreekanth, D. Sunitha · 2013
In speech processing applications often it is observed that many algorithms implemented so far in the past were able to concentrate either on reducing the noise or improving the speech intelligibility, but not the both. The algorithm introduced in this paper focuses on reducing the noise in the speech signal while improving its intelligibility. The new algorithm is based on probabilistic synthesis and analysis of speech signal. Key words: speech intelligibility, synthesis, baysian probability, binary mask. Gaussian mixture models (GMMs) to be used as classifiers. Figure1.1 shows the general block diagram of algorithm. In noisy environments, the speech signal SNR is very low and negative some times. Algorithms that improve speech quality do not necessarily improve speech intelligibility. This is most likely due to the distortions introduced to the speech signal. In contrast to speech quality, intelligibility relates to the understanding of the underlying message or content of the spoken words, and is often measured by counting the number of words identified correctly by human listeners. Intelligibility can potentially be improved only by suppressing the background noise without distorting the underlying target speech signal. 1.