Enhancement of speech intelligibility and naturalness using radwt

Abstract

Speech plays a major role in the effective communication between humans. owever, laryngectomees suffer from the loss of their vocal tract which is vital for speech production, due to various speech disorders such as Laryngeal cancer, surgeries etc. Hence, the need for speech assistive devices arose which initially started with text enabled devices where the patients were not able to communicate orally in a limited form. There were different forms of alternative assistive systems such as esophageal speech, Electro larynx and Tracheoesophageal speech. The assistive devices then evolved with a simple electromagnetic transducer which was able to produce a monotonic and robotic voice. Patients who lose their speech due to the loss of the vocal tract adapt one of the three alternative forms of speech production such as Electrolaryngeal speech produced by assistive devices such as Electro larynx, Esophageal speech without any external device and the Tracheo-esophageal speech using an intra-oral device implanted into the throat. In all the three cases speech is intermittent and produces less intelligible speech that is not similar to natural speech. Hence, it was proposed to enhance the speech intelligibility and naturalness of the produced speech using advanced techniques to provide comfort of listening without much strain to the listener as well as the speaker. The observations were collected from subjects of different ages, regional language and gender to measure the intelligibility and naturalness of the uttered speech. The intelligibility of speech is defined as the ability of the listener to discern the speech with reference to the recognition of the word independent of its meaning and the intelligibility of all the forms of speech is found to be as follows, newlineTable A0. Mean Opinion Score (MOS) based on Intelligibility Type of alternative speech Environment MOS newlineElectrolaryngeal speech In the presence of white noise and musical noise 40.6% newlineEsophageal speech 50% newlineEnhanced speech using RADWT 85% newline newline