Abstract

In this paper, we have taken Electro-Larynx (EL) speech and have improved the speech quality, electro-larynx speech was improved in terms of naturalness and intelligibility by introducing variations in the F0-contour and template matching with correlation coefficient. Initially, we introduced two different speech signals, the first speech signal introduced was healthy speech signal and the second speech signal introduced was disordered speech signal. Here, the second speech signal, the disordered speech is taken as the EL speech. The fundamental frequency or pitch was extracted first from the two inputed speech signals, then the contour of each fundamental frequency was extracted from the two input speech signals. Using these extracted features of fundamental frequency the gender classification by K-means algorithm was instigated. The same process was implemented with F0 contour features which was extracted using K-NN algorithm. EL speech contains directly radiated electrolarynx noise (DREL). The noise was filtered out using spectral subtraction algorithm. Once DREL noise is removed from EL speech, the quality of the speech was greatly improved. Then EL enhanced speech signal is compared and mapped with healthy speech signal using template matching algorithm with the help of correlation coefficient, this improves the overall quality, that is the naturalness and intelligibity of the introduced disordered speech signal. This technique helps solve the major problem of speech faced by differently abled persons with larynx disorder.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call