Abstract
A subjective experiment is described which confirms that characteristics of the spectral envelope of speech, notably formant positions, contribute to our ability to recognise a speaker's voice. Speakers with similar accent and rate of speaking were selected and their larynx frequency characteristics were eliminated in a vocoder system with fixed voicing frequency. A direct subjective comparison of resulting voice samples showed that the speakers' voices were still highly recognisable. A proportion of the voice samples were subjected to further treatment consisting of constant shifts of formants. It was found that the recognisability of the samples was destroyed although intelligibility was unimpaired.
Get full access to this article
View all access options for this article.