From contraction of muscles to sound, a model based study

Despite several decades of speech production research, the process of transforming the speaker’s intended message as formulated in the brain to an acoustic output is not fully understood. One of the main bottlenecks is measurement. Speech production is a process which starts in the brain by the speaker’s intention to say something and ends with an acoustic signal as the final product. There are several blocks which are involved in the process and are extremely difficult or near impossible to measure these blocks accurately and precisely with available technology. An alternative way is to use computer models to solve an inverse problem. That is, for a given output (e.g. acoustic signal or articulation trajectory), we estimate the input signal (e.g. contraction of muscles). In general, solution to an inverse problem is not unique. So, we need some criteria to decide which solution to choose. This is usually done either by optimizing a cost function or by applying some other knowledge. In this seminar, I’ll present our work of using computer models to investigate acoustic, articulatory and muscles contraction during production of vowels or vowel-vowel sequences.