Conversion of Non-Audible Murmur to Normal Speech Based on FR-GMM using Non-Parallel Training Adaptation Method

In recent years, majority of people are affected with low voice, due to the biological variations in the environment and hereditary infections occurred in throat or vocal track. These people communicate with others in a murmured low voice utterances. This paper introduces a novel approach for the processing of non-audible murmur (NAM), with respect to a Full Rank Gaussian Mixture Model (FR-GMM) using non-parallel training adaptation method. Non-audible murmur means, lightly uttered soft voice that is very difficult to hear. The non-audible murmur is extracted directly from the soft tissues behind the ear, utilizing a body conductive non-audible murmur microphone made of stethoscope sensor. In the proposed technique, a FR-GMM is prepared with the traditional strategy while allowing the reference speaker voices using nonparallel training adaptation method. The Investigational analysis outputs prove that the proposed approach improving the natural quality of speech by 6% in word precision compared to the traditional approaches.