Paper: Noise as a Tool for Spoken Language Identification

ACL ID I08-2114
Title Noise as a Tool for Spoken Language Identification
Venue International Joint Conference on Natural Language Processing
Session Main Conference
Year 2008

Segmental SNR (Signal to Noise Ratio) is considered to be a reasonable measure of perceptual quality of speech. However it only reflects the distortion in time dependent contour of the signal due to noise. Objective Measures such as Log Area Ratio (LAR), Itakura-Saitio Distortion (IS), Log-Likelihood Ratio (LLR) and Weighted Spectral Slope (WSS) are better measures of perceptual speech quality as they represent deviation in the spectrum. Noise affects the speech time contour and the corresponding frequency content. Different languages have some peculiar characteristics due to variation in the phonetic content and their distribution. Distortion introduced by noise and application of enhancement algorithm varies for different phonemes. In this paper a novel idea of ...