The National Science Council subsidizes the Academia Sinica project. Cao Yu (right 1), deputy director of the Academia Sinica’s Information Technology Innovation Research Center, and his team spent 9 years building an AI deep learning assisted voice system, which can eliminate background noise during communication and improve Speech abnormalities caused by electronic larynx or oral cancer make communication more "loving".

(Photo by reporter Wu Boxuan)

[Reporter Wu Baixuan/Taipei Report] The National Science Council promotes artificial intelligence (AI) forward-looking technology, and subsidizes the team of Cao Yu, deputy director of the Information Technology Innovation Research Center of the Academia Sinica, to develop a "speech enhancement system based on deep neural networks" to improve communication and understanding The listening part can eliminate background noise, and the speaking part can strengthen the slurred speech of oral cancer patients. There are already APPs and the industry has developed artificial electronic ears and other assistive devices. Currently, it is locked in Chinese, and it will not rule out applying to various languages ​​in the future.

Cao Yu, who developed the technology, said that the team started investing more than 9 years ago, using AI deep learning to enhance the speech system, which is specially optimized for "understanding", divided into "listening to AI" and "speaking AI". Listening to AI will distort speech Enhanced with a deep neural network, the AI ​​is trained to achieve a clean voice effect and completely eliminate background noise such as traffic and music.

Please read on...

Speaking of the AI ​​part, speaking dysfunction is considered, and abnormal voice detection and abnormal voice enhancement are adopted. For example, patients with oral cancer or users of electronic larynx can make the original damaged voice clearer, and then play it back with mobile phones or instruments.

In addition, listening to AI can completely remove background noise. Cao Yu said that through the cooperation of the industry, special processing is also carried out for special events such as boiling water, alarm sounds, etc., so that in the future, the auxiliary listening system will not only enhance the voice, but also have different warning detection. .

Zeng Yu said that the team is the first research in the world to apply deep learning to voice-enhanced artificial electronic ears. The results won the National Innovation Award and the Future Technology Award. Currently, the research results are public. During the implementation of the technology transfer project, the current technology development language is mainly Chinese, and it can be extended to various languages ​​in the future with minor adjustments.

Lin Guanghong, Secretary-General of the National Science Council, said that the Academia Sinica has specially invested AI technology in the development of novel articulation aids for people with related needs, such as oral cancer patients, electronic larynx, etc. When the pronunciation is not so clear, the speech quality and recognition can be enhanced through technology , improve the ability to communicate with others, and bring a first-line "sound" machine.

Cao Yu, deputy director of the Information Technology Innovation Research Center of Academia Sinica, and his team developed AI technology, using deep learning to enhance the speech model, helping to improve listening and speaking, such as distortion and background noise elimination, or making the pronunciation of electronic throat users easier to understand .

(Photo by reporter Wu Boxuan)