A series of experiments tested the contribution of phase information to the naturalness of residual excited linear prediction (RELP) coded speech, that is, how close a speaker's voice sounds under RELP coding to its unprocessed analog. Listeners were presented with a corpus of sentences uttered by eight speakers (four male and four female). Three classes of approximations of the residual phase spectra were employed on these sentences: quantized approximations for decreasing bit rates, noise-corrupted approximations for increasing noise levels, and combinations of these over varying window sizes. Listeners classified these utterances on the basis of their proximity to unprocessed speech. Results indicate that residual phase spectrum provides information about voice timbre and quality and acts as a cue to speaker identity in coded speech.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.