Using computer-based speech recognition technology as a practice partner for persons with severe motor speech disorders.

摘要

An extensive amount of practice is required for traumatically brain injured persons to improve their dysarthric speech. The present research project investigated the role(s) of computer technology in practice strategies for these speakers. The overall aim of the research program was to examine the impact that computer speech recognition technology has on the sound production (articulation) performance of adults with motor speech disorders. An experiment was designed to demonstrate the relationship between human listeners and computer-based evaluations of speech production and to examine the phonetic intelligibility patterns for individuals with dysarthria secondary to traumatic brain injury. Recordings were made of four dysarthric speakers producing repetitions of individually selected target stimulus sounds. For two dysarthric speakers, the intended stimuli consisted of multiple productions with different prevocalic consonant variation in a /Ci/ context. For two dysarthric speakers, the intended stimuli consisted of multiple productions with different vowel variation in an /hV/ context. Computer-based speech recognition system ratings were made by interfacing a video camera recorder with the speech recognition system. Five trained listeners and five untrained listeners were asked to listen to recorded samples and to identify the randomly produced repetitions of sound combinations for each dysarthric speaker from a list. Phonetic intelligibility accuracy of articulatory performance and the patterns of phonetic intelligibility were measured for each dysarthric speaker and subjected to statistical tests for significance. One hypothesis was that phonetic intelligibility accuracy was similar among trained listeners, untrained listeners, and speech recognitions system accuracy performance evaluations when comparing the intended sound productions with the perceived sound productions. In addition, the degree of agreement for phonetic intelligibility patterns across trained listeners, untrained listeners, and speech recognition system evaluations was hypothesized to be similar when comparing the perceived sound productions. The results revealed complex interrelationships of phonetic intelligibility accuracy and recognition patterns across the dysarthric speakers. Relationships related to phonetic intelligibility accuracy were observed across human listeners and speech recognition system evaluations for specific dysarthric speakers. Phonetic intelligibility agreement between human listeners and the speech recognition system differed for specific dysarthric speakers, but not for others.