JapanEEG: a 1000-hour EEG–EMG–audio dataset of Japanese speech production
Year: 2026
Type: dataset
Abstract: JapanEEG contains 1020 hours of simultaneously recorded scalp electroencephalography (EEG), facial electromyography (EMG) and speech information from three healthy native Japanese speakers during open-vocabulary overt speech, with additional covert-speech and listening sessions. Recordings were made with three EEG systems (g.Pangolin, 128 channels; g.SCARABEO, 62 channels; eego, 63 channels) over many sessions spanning several months, and are organized according to the Brain Imaging Data Structure (BIDS). The release includes EEG, EMG, speech-event annotations, automatic transcriptions, and word and phone timing. Bad EEG channels are marked in the channel files for continuous recordings and in the event files for shuffled fragments where available. Released speech audio is voice-masked using Seed-VC with a fictional reference voice synthesized using Qwen3-TTS. For continuous listening to commercial audiobooks, the release provides a 30 Hz low-pass amplitude envelope, automatic transcriptions, and source references. The original audiobook audio is excluded. For the listening/covert repetition task, automatic transcriptions and voice-masked reference audio are provided where available. Readings of copyrighted texts are released as utterance-level segments in random order (ses-shuffled). Araya Inc. hosts the tar.xz archives that extract to the BIDS tree (see README). To request access, sign in to Harvard Dataverse, provide your name, affiliation, institutional email address, and intended use, and agree to the JapanEEG Data Use Agreement in the Terms tab. Approved users receive credentials for the Araya download server.
Source: Harvard Dataverse
DOI: 10.7910/dvn/5rm3ot
Institution: University of California, San Francisco
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7225786113
Cites:
Cited by:
Related to: