Effects of selected vocal disguises upon speaker identification by listening

Abstract

This research was designed to investigate the effects of selected vocal disguises upon speaker identification by listening. The experiment consisted of 360 pair discriminations presented in a fixed-sequence mode. The listeners were asked to decide whether two sentences were uttered by the same or different speakers and to rate their degree of confidence in each decision. The speakers produced two sentence sets utilizing their normal speaking mode and five selected disguises. One member of each stimulus pair in the listening task was always an undisguised speech sample; the other member was either disguised or undisguised. Two listener groups were trained for the task: a naive group of 24 undergraduate students, and a sophisticated group of three doctoral students and three professors of Speech and Hearing Sciences. Both groups of listeners were able to discriminate speakers with a moderately high degree of accuracy (92% correct) when both members of the stimulus pair were undisguised. The inclusion of a disguised speech sample in the stimulus pair significantly interfered with listener performance (59%--81% correct depending upon the particular disguise). These results present a similar pattern to this authors' previous results utilizing spectrographic speaker-identification tasks (Reich et al., 1976).

Voice disguise and automatic speaker recognition

“…Even for nasals the effect is greater but only limited to F1 and F2. This is coincident with Reich and Duke's investigation in 1979 [9]. Therefore the degradation resulted from pinched nostril for speaker recognition seems not as great as we expected and measured both acoustically and automatically though the auditory voice quality is greatly changed.…”

“…He found that the accuracy of speaker identification by machine matching with disguised voices is only ''little better than chance'' [4][5][6]. Using spectrogram analysis, samples containing disguised voices will obviously lower the correct rate of speaker identification and affect the error rates, especially the increase of identification error and the proportional decrease of exclusion [7][8][9] was decreased by 13% after the interval of 1 day [10]. Many studies have showed significant negative influences of voice disguise on speaker identification.…”

Detecting the presence of vocal disguise in the male voice

“…If one accepts the assumption made by Tosi (1979 1976; Reich and Duke, 1979). Two of six speaking conditions reported previously were used in the present experiment.…”

“…In a study of fundamental frequency disguise (McGehee, 1937), correct recognition was reduced by 13% for a one-day time interval. Reich and Duke (1979) investigated the effects of extemporaneous and rehearsed vocal disguises in a pair-discrimination listening task. The inclusion of a disguised speech sample in the stimulus pair significantly interfered with listener performance (59%-81% correct depending on the particular disguise).…”

Auditory discrimination of natural speech and synthetic speech used as voice disguise

“…Impostors may change phonation by using glottal fry or whisper voice; change pitch; change accents by using a foreign accent or another dialect; or do anything else in their power to hide their identities. Especially in forensics, the importance of awareness of voice disguise has been emphasised since the 1970s [1][2][3][4][5][6][7][8][9][10]. In the last two decades, researchers have warned of a new type of voice disguise, the use of synthetic speech [11][12][13].…”