Refining Speech Clarity with Wavelet Denoising under Different Face Mask Conditions: A Subjective Analysis

Amid the COVID-19 pandemic, people have adopted various face masks and face shields as protective measures against infection. While these measures have been instrumental in saving countless lives, they pose significant challenges to interpersonal communication, especially in scenarios requiring clear verbal interaction. This study utilizes a microphone to capture speech signals in different scenarios involving face masks, with and without face shields. Participants read vowels and the Grandfather Passage across ten experimental conditions, including surgical masks, cloth masks, double masks (surgical and cloth combination), and N95 masks, both with and without face shields. The obtained speech signals, often distorted by noise and reverberation, undergo enhancement through the wavelet denoising approach using discrete wavelet transform with soft thresholding. The quality of the enhanced signals was compared to the original acquired signals using a subjective comparison test involving 30 listeners who rated the signals based on comparison mean opinion scores (CMOS). Multiple research findings indicate that the signal improvement achieved through wavelet denoising consistently exceeds the quality of the initial signal, even under challenging conditions such as double masks with face shields. This study highlights the practical efficacy of wavelet denoising in addressing speech clarity challenges caused by protective face coverings, offering a valuable solution for improved communication in masked environments.