LibriBrain
收藏资源简介:
LibriBrain数据集是一个非侵入性的MEG数据集,由一位健康参与者聆听超过50小时的音频书籍所采集。该数据集包含超过50小时的MEG记录,通过306个传感器覆盖整个头部/大脑。数据集与LibriSpeech语料库中的单词和音素级别对齐,便于自动语音识别(ASR)。数据集被分成训练、验证和测试集,并包含额外的比赛保留集用于排行榜更新和最终排名。LibriBrain数据集的发布旨在推动非侵入性脑机接口的进步,特别是在语音解码方面。
The LibriBrain dataset is a non-invasive magnetoencephalography (MEG) dataset collected from a healthy participant while they listened to over 50 hours of audiobooks. It contains more than 50 hours of MEG recordings that cover the entire head and brain via 306 sensors. The dataset is aligned with word and phoneme-level annotations from the LibriSpeech corpus, supporting tasks related to automatic speech recognition (ASR). It is split into training, validation, and test sets, and also includes an additional competition holdout set for leaderboard updates and final ranking. The release of the LibriBrain dataset aims to advance the development of non-invasive brain-computer interfaces (BCIs), particularly in the field of speech decoding.

- 1The 2025 PNPL Competition: Speech Detection and Phoneme Classification in the LibriBrain DatasetPNPL · 2025年



