Exploring muscle activation patterns in speech

Dnr:

SNIC 2016/1-525

Type:

SNAC Medium

Principal Investigator:

Olov Engwall

Affiliation:

Kungliga Tekniska högskolan

Start Date:

2016-12-27

End Date:

2018-01-01

Primary Classification:

10208: Språkteknologi (språkvetenskaplig databehandling)

Secondary Classification:

10610: Bioinformatik och systembiologi (metodutveckling under 10203)

Webpage:

Allocation

Abstract

The aim of the project is to investigate if muscle activation patterns (MAP) in speech are generated to minimize required energy or variability. Motivation: Generation of speech requires co-ordinated activation of tens of different muscles to control the speech organs. Since the acoustic output is determined by the shape of the resonance tube; the vocal tract; and the same shape could be generated by many different MAPs, an important, but so far unanswered, question may be raised: How does the brain choose MAP for a specific speech task? An influential hypothesis argues that speech gestures are selected based on motor “economy” and therefore, the optimal MAP would be the one requiring minimum energy. However, experimental data from hand movements suggest that, instead of energy, animals tend to optimize MAP to minimize the variability of the movements. Indeed reliable movements despite noise in the brain, muscles and environment provide an evolutionary advantage. Consistent with this, even the brain activity becomes less variable during sensory and motor processing. This project explores these two hypotheses, in particular with respect to how different speech contexts influence the MAPs for the tongue and jaw muscles. In order to study these hypotheses, vast combinatorial searches of MAPs, resulting vocal tract tubes and acoustic resonances need to be performed in simulations. These simulations require HPC resources to be feasible. The simulations were initiated under the SNIC 2015/1-423 project with Sten Ternström as Principal Investigator.