Least-squares mapping from kinematic data to acoustic synthesis parameters for rehabilitative acoustic learning.

摘要

Thousands of people suffer from dysarthria resulting from neurological injury of the motor component of the motor-speech system, and need to rely on alternative methods to communicate in daily life, such as body language or text-to-speech. However, there are currently very few effective rehabilitative therapies for helping these patients improve their speech. Because of this, research is needed to develop better rehabilitative therapies. One such area of research is the use of involuntary acoustic learning. The Speech and Swallowing lab at Marquette University has an Electromagnetic Articulography (EMA) system to collect kinematic data and a software system called Rehabilitative Articulatory Speech Synthesizer (RASS) that is able to create the necessary synthesized acoustic feedback to study the effects of these kind of therapies.;One key aspect of the RASS system is the mapping from kinematic sensor data to acoustic synthesis parameters. This is a complex problem that depends on individual subject anatomy and vocal tract patterns. Currently, the RASS system uses a simple piecewise linear method, but it would be advantageous to improve this to be more accurate across a wider range of vocal configurations. The goal of the research work presented here is to develop and test new approaches for kinematic to synthesis mapping, in the hopes of improving the quality and intelligibility of the RASS system.;Results indicate that the new mapping gives reduced mapping error. Ultimately, the impact of this work is that it provides researchers with a more accurate method for mapping kinematic data to synthesis parameters.