Noise management in mobile speech based health tools

Noise impacts speech recognition and processing capabilities on mobile devices. The Signal-to-Noise Ratio (SNR) is a good indicator of noise in the acoustic environment where the speech is recorded and processed. In this paper, SNR and its impact on speech recognition and processing capabilities on a mobile device are studied for a relatively small text corpus of 50 words. The speech recognition accuracy is quantified using a word accuracy metric for different levels of SNR. Future mobile health tools that detect speech disorders caused by illness can benefit from this study and the resulting tool developed to give appropriate feedback to a user about their acoustic environment. Corrective measures can be suggested based on this to alleviate the potential problem, e.g., adjusting a microphone or relocating to a quieter environment.