Electrolaryngeal Speech Identification using GMM
R. Nandana · International Journal of Engineering Research and · 2020
People who have lost their larynx due to laryngeal cancer or due to other physical conditions cannot produce voice like other humans.As air is breathed out through the vocal folds, vocal folds are vibrated and sound is produced, heard as a speech voice.In situations where larynx is removed, air can never again go from the lungs into the mouth.The association between the windpipe and the mouth never again exists.Electrolarynx is a battery driven machine that produces sound to create a voice for laryngectomy patients.The electrolarynx could either function indirectly through contact with skin, that triggers pharyngeal vibrations or specifically by intraoral pressure, which produces vibrations of the vocal cavity.Articulation muscles are usually intact after total laryngectomy(TL) and therefore capable of transforming the stimulus noise supplied into understandable voice.Traditional electrolarynx produces a robotic voice and a mechanical humming when used.This paper focuses on increasing the quality of electrolaryngeal speech.Here, implementation of Mel Frequency Cepstral Coefficient (MFCC) Algorithm and Gaussian Mixture Model (GMM) in pair is used to achieve the target.We have considered MFCC with "tuned parameters" as the primary feature and delta-MFCC as secondary feature.And, we have implemented GMM with some tuned parameters to train our model.Speech with highest score is identified and corresponding normal speech is produced using python platform.