Investigation on optimum parameters for LPC based vocal tract shape estimation

Hearing impaired individuals having severe or profound deafness are not able to speak due to the lack of auditory feedback. Commercial speech training aids giving feedback of acoustic and articulatory parameters are available for improving the speech produced by the hearing impaired persons in terms of intonation, stress, pronunciation, and articulation accuracy. Aids based on visual feedback of vocal tract shape are reported to be useful for the improvement in speech production. Investigations for vocal tract shape (VTS) estimation based on LPC analysis of speech, as reported by Wakita, are reported in this paper. In Wakita´s algorithm VTS is obtained by appropriately selecting some of the parameters including vocal tract length, LPC order, and speech sampling rate. The objective of our investigation is to optimize these parameters for realistic estimation of VTS. The results are obtained for recorded vowels from adults as well as children from various age groups. The results for adults are verified based on available magnetic resonance imaging (MRI) data. Due to non-availability of MRI images for VTS of children, validation of our results for children is based on overall growth of vocal tract as reported in the literature and also by comparing with the results from researchers who have used other indirect techniques.