Impact of ERB and bark scales on perceptual distortion based near-end speech enhancement
G. V. Nikhil, A. M. Keerthi, B. S. Premananda · 2017 International Conference on Trends in Electronics and Informatics (ICEI) · 2017
Speech is the vocalized form of communication. Exchange of information among individuals happens through speech. Speech degradation can occur due to the ambient noises at the speaker end and listener end. At the speaker end, the microphone in the mobile capture both the speaker's speech signal as well as the noise, therefore the influence of background noise at speaker end can be reduced using filtering or using some of the noise reduction algorithms. But at the listener end, the noise directly reaches the listener's ear therefore it is impossible to remove the influence of the noise. Therefore pre-processing of speech is performed, which can be done in frequency or time domain. Among the two, frequency domain is most preferred implementation, as it is more realistic with reference to human audibility. Perceptual Distortion is the measure of the amount of distortion of the speech signal in the presence of the background noise. In the proposed work, pre-processing in frequency domain algorithm considering perceptual distortion is used to improve the intelligibility of the speech for the near-end listener, in the presence of background noise at the listener end. The gain for the speech enhancement is derived based on perceptual distortion measure. The speech enhancement technique is performed for both Bark and ERB scale and the results are analyzed.