Evaluating Speech Intelligibility with Processed Sound
Loading...
Date
2023-09-18
Authors
Faulkner, Sam
Journal Title
Journal ISSN
Volume Title
Publisher
University of Waterloo
Abstract
This paper was created with the goal of researching the different impacts that background noise can have on listeners' ability to interpret speech. The brain is responsible for separating speech and noise, but this can be difficult if this organ is damaged or the noise is too overwhelming to separate out. I partnered with Augmented Heating.io to see whether their noise reduction software can do some of this processing on behalf of the brain. This would reduce cognitive effort and help make conversations more accessible in noisy environments. To research this topic, I created a study that evaluated participants' ability to understand words that have often confused sounds in them. These words were presented with different types of voices, with different kinds of background noise, and both with and without processing from AugmentedHearing's algorithms.
Preliminary results indicate that intelligibility scores were not higher for the denoised speech compared to the noisy speech. This was not the expected result, however, there is still much to consider within the data. These preliminary findings are grounds for further studies and will hopefully lead to an improvement in future iterations of the speech processing software.
Description
Keywords
psychology, artificial intelligence, sound processing, noise removal