Wavelet Transform based Enhanced Statistical Voice Conversion of Non-Audible Murmur (NAM) for Improved Speech Intelligibility

This article offers a novel wavelet transformation based statistical approach for transforming Non-Audible Murmur (NAM) signals into speech sounds in order to overcome the problems arising from the low energy and low-frequency content of NAMs. The wavelet transform of the NAM signals safely partitions the signals into detailed time-frequency features, enabling both noise reduction and feature enhancement. Some of the proposed system design steps involve pre-processing, framing, wavelet analysis of the underlying signal, and reconstruction of the enhanced signal. The simulation results show that this method can effectively reduce the standard deviation of the converted speech to 98% of the word accuracy rate. This shows that the wavelet transform is beneficial in enhancing the quality of the speech by filtering out noise and making it easier to implement in any application where clear speech reproduction is crucial.