Mobile microphone robust acoustic feature identification using coefficient of variance

Abstract One of the most challenging techniques for speech analysis applications in mobile phones is acoustic feature extraction. The adverse environment noises, diversity microphone specifications, and various recording software have a significant effect on values extracted features. In this study, we investigate robustness different types features related to time-based, frequency-based, sustained vowel using 11 devices. 49 recordings subjects reciting Rainbow Passage 25 /a/ were collected. By way synchronous recording, analyzed compared 253-dimensional vectors order examine how consistent data between variability was measured method coefficient variance. Data with low identified be from such as transition parameters, amplitude modulation, contrast, Chroma, mean fundamental frequency formants. These groups turn out more reliable than others their dependency device specifications.