Title :
Non-intrusive speech intelligibility assessment
Author :
Sharma, Divya ; Naylor, Patrick A. ; Brookes, Mike
Author_Institution :
Nuance Commun. Inc., Marlow, UK
Abstract :
We present NISI, a novel non-intrusive speech intelligibility assessment method based on feature extraction and a binary tree regression model. A training method using the intrusive STOI method to automatically label large quantities of speech data is presented and utilized. Our method is shown to predict speech intelligibility with an RMS error of 0.08 STOI on a test database of noisy speech.
Keywords :
feature extraction; mean square error methods; speech intelligibility; trees (mathematics); NISI method; RMS error; binary tree regression; feature extraction; noisy speech; nonintrusive speech intelligibility assessment; speech data; training method; Abstracts; Covariance matrices; Electronic mail; Measurement; Nickel; Silicides; Speech; Classification and Regression Trees; Data Driven; Speech Intelligibility; Speech Quality;
Conference_Titel :
Signal Processing Conference (EUSIPCO), 2013 Proceedings of the 21st European
Conference_Location :
Marrakech
