Speaker Identification Based on Nasal Phonation

Abstract

A method of automatic speaker identification based on the physiology of the vocal apparatus and essentially independent of the spoken message has been developed. Power spectra produced during nasal phonation are transformed and statistically matched. In experiments involving the identification of individual speakers out of a population of 10 speakers, an average identification accuracy of 97% was obtained. With an experimental population of 30 speakers, identification accuracy was 93%. The procedure is outlined, and the experimental evaluation of it is described.

Perception of the [m]–[n] distinction in CV syllables

“…For these reasons, [ m (•) ] may be expected to contain the most salient place of articulation cues, followed by [ m (u) ] and [ n ] murmurs, while the elevated tongue body during [ m (i) ] may in fact make this murmur more [ n ] -like than the [n] murmurs.The data for uncontaminated isolated murmurs (cut at -1, -2, -3) were submitted to ANOVAs, which yieldIndividual syllable scores in the murmurs condition. the lower scores for I-(i)] murmurs; a consonant by vowel interaction[F(2,22)-13.45, p --0.0002; F(2,8) = 4.76, p --0.0435], reflecting the presence of a vowel effect for [ m ] but not for [ n ] murmurs; and a consonant by duration interaction [F(2,22) --6.31, p ----0.0068; F(2,8) --5.00, p < 0.0389 ], which apparently derives from the fact that [ n ] murmurs, but not [ m ] murmurs, suffered from the excision of the penultimate pitch pulse (cut at-1 vs -2).8 The lower F values in the AN-OVA across talkers indicate considerable talker variability in nasal murmur spectra, a well-known phenomenon often commented on in the literature (e.g.,Fant, 1960;Fujimura, 1962;Glenn and Kleiner, 1968). The unpredictable nature of that variability, as compared to the somewhat more regular scaling differences for oral resonances, may also have been responsible for the overall difference in scores between isolated murmurs and vowels in the present mixed-talker design.…”

Gender Differences in Long-Term Average Spectra of Children's Singing Voices

“…35,78 It is also possible that some of these parameters may be in the nature of interaction traits, comprising both physical and behavioral elements that impinge on habitual oral cavity and VT configurations. 30,79,80 It is also possible that the information used by listeners may not be consistent from one speaker/singer to another, and listeners themselves may differ. The best we can do at present is to examine the freedoms and limitations of several possible parameters; to attempt to identify a single criterion factor is possibly fruitless.…”

Formant dynamics of bilingual identical twins

“…Garvin & Ladefoged, 1963;Glenn & Kleiner, 1968;Wolf, 1972) claims that any type of between-speaker difference is either organic (i.e., caused by different anatomical structures of the vocal tract) or learned (caused by different articulatory manners acquired by speakers). In this sense, identical twins who stay together would have identical voices, as they are identical in anatomy and also share the same language environment.…”