EEG and Audio for Auditory Attention and Speech-in-Noise Comprehension

Octave Etard's dataset contains unprocessed 64-channel EEG recordings from 18 participants listening to audiobooks under various listening conditions. The data includes native (English) and foreign (Dutch) language speech presented in quiet, speech-shaped noise, and competing-speaker scenarios, with stimuli and precise time-alignment annotations provided. This multimodal resource supports research into neural speech tracking, auditory attention decoding, and the brain's response to degraded speech.

Use Cases

• Decoding selective auditory attention based on EEG responses to competing speakers.
• Investigating neural correlates of speech comprehension in noise using the provided SNR levels.
• Developing models for neural speech tracking in theta and delta frequency bands.
• Benchmarking EEG preprocessing and linear modeling techniques for auditory brainstem responses.

Strengths

• Includes precise stimulus-onset timestamps stored as annotations, enabling accurate EEG-audio alignment.
• Contains data from 18 participants across multiple controlled listening conditions, including a foreign language (Dutch) subset for 12 participants.
• Provides raw data in multiple formats (VHDR, HDF5, CND-like) for flexibility in analysis.

Limitations

• Key metadata such as exact row count, file size, and last update date are unspecified across all sources.
• Analysis requires careful cropping of specific time segments (e.g., first/last seconds of noise trials, embedded English sentences) as detailed in the notes.

Provenance

Source

Octave Etard, Imperial College London

Collection Method

EEG recorded at 1 kHz via actiCHamp amplifier with electrodes in the 10-20 system; audio recorded simultaneously via StimTrack device for alignment.