FEIS: Fourteen-Channel EEG Recordings of Imagined and Spoken Speech

FEIS is a multimodal dataset containing electroencephalography (EEG) recordings from 23 participants using an Emotiv EPOC+ headset. It captures brain signals during listening, imagined speech, and actual speaking of 16 English phonemes and 16 Chinese syllables. The dataset includes raw EEG data, participant audio recordings, and the complete experimental setup for replicability.

Use Cases

• Classifying imagined phonemes based on time-series EEG signals.
• Comparing neural activity patterns between heard, imagined, and spoken speech phases.
• Developing multilingual speech decoding models using data from English and Chinese stimuli.
• Benchmarking lightweight, mobile EEG headsets for brain-computer interface applications.

Strengths

• Includes data from 23 participants, providing a foundation for multi-subject analysis.
• Captures a complete experimental pipeline with raw data, audio, visual prompts, and processing scripts, supporting full replicability.
• Contains multimodal recordings (EEG and audio) across three distinct cognitive states: listening, imagining, and speaking.

Limitations

• Key metadata such as row count, file size, and last update date are unknown across all sources.
• The participant pool is imbalanced, with 21 participants for English phonemes and only 2 for Chinese syllables.
• The dataset relies on a low-density, 14-channel mobile EEG headset, which may limit spatial resolution compared to research-grade systems.

Provenance

Source

University of Edinburgh, from MSc dissertations by J. Clayton and S. Wellington.

Collection Method

Recorded using an Emotiv EPOC+ EEG headset in a controlled experiment with audio-visual prompts.

Time Range

2019 (inferred from dissertation submission years)