On Short-Time Estimation of Vocal Tract Length from Formant Frequencies

Abstract

Vocal tract length is highly variable across speakers and determines many aspects of the acoustic speech signal, making it an essential parameter to consider for explaining behavioral variability. A method for accurate estimation of vocal tract length from formant frequencies would afford normalization of interspeaker variability and facilitate acoustic comparisons across speakers. A framework for considering estimation methods is developed from the basic principles of vocal tract acoustics, and an estimation method is proposed that follows naturally from this framework. The proposed method is evaluated using acoustic characteristics of simulated vocal tracts ranging from 14 to 19 cm in length, as well as real-time magnetic resonance imaging data with synchronous audio from five speakers whose vocal tracts range from 14.5 to 18.0 cm in length. Evaluations show improvements in accuracy over previously proposed methods, with 0.631 and 1.277 cm root mean square error on simulated and human speech data, respectively. Empirical results show that the effectiveness of the proposed method is based on emphasizing higher formant frequencies, which seem less affected by speech articulation. Theoretical predictions of formant sensitivity reinforce this empirical finding. Moreover, theoretical insights are explained regarding the reason for differences in formant sensitivity.

Influence of Speaker-Specific Parameters on Speech Separation Systems

“…To estimate the VTL we make use of methods described in [16] which are based on formant measurements and a physical model of the vocal tract. For voiced speech frames, we can get an estimate of the vocal tract length with the help of the lowest resonance frequency of a lossless uniform vocal tract Φ, which depends on the n-th formant Fn as…”

“…In practice, we consider the first four formants (N = 4) and set βn as proposed in [16] and shown in Table 1.…”

Vowel Space as a Tool to Evaluate Articulation Problems

“…Also, differences between male and female speakers were considerable. These observations suggest that vowel detection could be made more robust (e.g., [3,22]) and the normalization between male and female voices might be improved (e.g., [23,24]).…”

Estimation of the Vocal Tract Length of Vowel Sounds Based on the Frequency of the Significant Spectral Valley

“…A statistical, data-driven approach was proposed by Kirlin [11]. Lammert and Narayanan [12] proposed a method to estimate VTL based on the deviations of formant data from those of a uniform tube. For a uniform tube, the n-th formant frequency, Fnu, is an odd integer (2n-1) multiple of the first formant frequency, F1u.…”

“…For natural vowels, the formant frequencies (Fn) differ significantly from Fnu, the formant frequencies of a uniform tube of the same length as that of speaker's VTL. Hence, Lammert and Narayanan [12] proposed a weighting function βn in addition to the factor 1/(2n-1), i.e., βnFn/(2n-1), whose mean value is an estimate of F1u and detetmined the optimum weights βn using a data-driven approach. The higher formants get a greater weightage.…”