| Paper Abstract and Keywords |
| Presentation |
2024-06-07 09:30
A pre-trained representation learning model can be used to decode speech from intracranial recordings
Shoya Murakami, Shuji Komeiji, Kai Shigemi (TUAT), Takumi Mitsuhashi, Yasushi Iimura, Hiroharu Suzuki, Hidenori Sugano (Juntendo Univ.), Koichi Shinoda (Tokyo Tech), Toshihisa Tanaka (TUAT) |
| Abstract |
(in Japanese) |
(See Japanese page) |
| (in English) |
Deep learning has been shown to be effective in decoding the content of a speaker's speech from recordings of brain activity using intracranial electrodes (intracranial electroencephalogram). However, deep learning requires a large amount of data, and collecting invasive data is not easy. In this paper, we hypothesize that by using a pre-trained public model as a feature extractor and inputting its output to a simple classifier, it is possible to decode the content of speech even with a small amount of training data. Results of an evaluation experiment with 15 participants showed that the proposed method outperformed the Transformer, a relatively complex model, in estimation accuracy for 10 participants. |
| Keyword |
(in Japanese) |
(See Japanese page) |
| (in English) |
brain-computer interface / electrocorticogram / self-supervised learning / / / / / |
| Reference Info. |
ITE Tech. Rep. |
| Paper # |
|
| Date of Issue |
|
| ISSN |
Online edition: ISSN 2424-1970 |
| Download PDF |
|