Recently, under the support of National Science and Technology Support Program, a new breakthrough was made by the team of Prof. Niu Haijun from the School of Biological Science and Medical Engineering together with Beijing Advanced Innovation Center for Biomedical Engineering and National Research Center for Rehabilitation Technical Aids on acoustic characteristics of the vocal output and tonal control of electrolarynx (EL) in Mandarin speech. Besides a supplement to the theory, they also developed medical equipment like wheel-controlled pitch-adjustable artificial EL and sliding-mode pitch-adjustable artificial EL (two types of speech therapy equipment, which belongs to the list of second category medical devices).

The research findings were published in various renowned journals including Medical & Biological Engineering & Computing and IEEE Transactions on Neural Systems and Rehabilitation Engineering (the top journal in rehabilitation science), with Dr. Wang Li being the first author and Prof. Niu being the corresponding author. They aregranted 11 invention patents and the copyright for a piece of software , and they set 1 national standard for EL. The artificial EL invented by the team not only passed the test of National Institutes for Food and Drug Control, but is also frequently on display at the recommendation of the Ministry of Science and Technology of the People’s Republic of China. In particular, the invention made a hit at the InnoTech Expo 2018 in Hong Kong.

The artificial EL and vocalization system

Based on the research mentioned above, the team then put forward the idea of utilizing deep learning technology to enhance mandarin EL speech, attempting to introduce the WaveNet technology into the identification and conversion of compensated sounds. However, due to the low pitch of EL, the rate of identification was comparatively low. Through the collection of patients peech data, in-depth study on the identification and conversion system of EL speech and the application of CTC technology, the team successfully enhanced the EL speech. The research was published recently in Journal of Speech, Language, and Hearing Research. The results were made under the joint efforts of the School of Biological Science and Medical Engineering of Beihang University and Beijing Advanced Innovation Center for Biomedical Engineering. Qian Zhaopeng, a PhD student form the School of Biological Science and Medical Engineering, is the first author and Prof. Niu is the corresponding author. The research takes advantage of deep learning technology to solve the problem of mandarin EL speech, providing a new research approach for enhancing the quality and understandability of EL speech and benefiting the patients who have lost their larynx greatly.

Speech identification framework of EL

Reported by Guo Meng

Reviewed by Han Huiyu

Edited by Jia Aiping

Translated by Xiong Ting