Study of time and frequency variability in pathological speech and error reduction methods for automatic speech recognition.

摘要: In this work, we study the variations in time and frequency domains inside a Spanish language corpus of speakers with nonpathological pathological speech. We show how speech has greater variability duration words than non-pathological speech, while domain that vowels confusability increases by 18%. The baseline experiments Automatic Speech Recognition (ASR) demonstrate causes loss performance ASR systems. To reduce impact use recent Vocal Tract Length Normalization (VTLN) system: MATE (augMented stAte space acousTic modEl), as way improving systems when dealing who suffer any kind pathology. Experiments 17.04% 11.19% WER reduction using respectively.