Enhanced Non-Audible Murmur Speech Prosthetics via Dynamic Time Warping and Adaptive Acceleration

Non-audible murmur (NAM) signals create unique challenges for speech signal processing due to their low signal-tonoise ratio and lack of vocal tract vibration. This work attempts to enhance NAM-to-speech conversion by introducing a new approach using Dynamic Time Warping (DTW) and Adaptive Linear Acceleration (ALA) with an outlier removal strategy. The suggested approach processes NAM signals through the initial use of an outlier detection mechanism to remove unwanted silence-speech pairs, followed by a DTW-based alignment to align NAM signals precisely to their respective voiced speech. ALA is subsequently used to dynamically adjust alignment rates for enhanced naturalness and intelligibility in the generated speech. Experimental tests illustrate that the novel methodology has superior alignment accuracy and clearness of speech over standard practice, with 88.57% and 94.53% success in eliminating and detecting outliers. This study will help progress silent speech communication with benefits mainly accruing to assistive speech prosthetics, voice rehabilitation, and covert communication systems.