EXPERIMENTS IN DYSARTHRIC SPEECH RECOGNITION USING ARTIFICIAL NEURAL NETWORKS

In this study, we investigated the use of artificial neural networks ( ANNs) to recognize dysarthric speech. Two multilayer neural networks w ere developed, trained, and tested using isolated words spoken by a dy sarthric speaker. One network had the fast Fourier transform (FFT) coe fficients as inputs, while the other network had the formant frequenci es as inputs. The effect of additional features in the input vector on the recognition rate was also observed. The recognition rate was eval uated against the intelligibility rating obtained by five human listen ers and also against the recognition rate of the Introvoice commercial speech-recognition system. Preliminary results demonstrated the abili ty of the developed networks to successfully recognize dysarthric spee ch despite its large variability. These networks clearly outperformed both the human listeners and the Introvoice commercial system.