Identification of voices in electronically disguised speech

Abstract

We describe an experiment on the effects of electronic vocal disguise on speaker identification. 36 listeners were trained to identify four voices. They then participated in listening tests in which their task was to identify speakers in a set of short stimuli. In the first test the stimuli were natural. For the second test the voices were artificially disguised. The fundamental frequency of the stimuli was modified using Sony SoundForge. As predicted, the rate of correct responses fell for the modified stimuli. The more extreme disguises (±8 semitones) yielded the lowest scores. The majority of listeners performed above chance level in all except the –8 semitone condition. The results showed significant effects for both listener and speaker, in line with previous studies of undisguised and naturally-disguised speaker identification. The variation suggests further research is required to assess the robustness of those forms of electronic disguise used to protect vulnerable witnesses.