Inverse problems in the analysis of vocal communication

Spoken language is a fundamental phenomenon in human culture and the most sophisticated means of communication. The carrier of spoken language, the speech sound, is extremely complex. The core of computational analysis and synthesis of naturalistic speech signals is an inverse problem called Glottal Inverse Filtering (GIF). The input of GIF is a recording of a vowel sound, and the output is the excitation signal at the vocal folds and a parametric representation of the vocal tract. This is a variant of blind deconvolution. Bayesian inversion and a Monte Carlo Markov Chain approach is discussed, and improved GIF results presented as compared to traditional methods. Another speech-related inverse problem is assessing vocal loading based on diffuse tomography. Electrical impedance tomography imaging of the vocal folds is also discussed.