Background

　Speech is the primary means of communication for human beings and plays a crucial role in maintaining one’s quality of life (QoL). However, patients with tongue resection and tongue movement disorders can only utter speech with low intelligibility because of a lack of, and inadequate control of speech articulators, respectively. According to the National Cancer Center Japan, the number of patients with cancer of the tongue was 22,800 in 2017, Japan.

Goal and Current Status

　To improve the intelligibility of speech uttered by the patients, prosthesis approaches that put wearable devises in their mouth have been the most popular. As a new approach, we adopted a machine learning in signal processing, particularly a voice conversion algorithm that reconstruct spectrum features by comparing those of patient‘s speech and healthy speaker’s speech. Because our approach does not require that an individual wear any special equipment, which results in more natural opportunities for speech communication. Listening evaluations showed improvement of speech intelligibility in phonemes of fricatives and stops.

URL

http://www.cc.okayama-u.ac.jp/~2hotetsu/DreamConversation/DreamConversationTOP.html