Kymata Soto Language Dataset: an electro-magnetoencephalographic dataset for natural speech processing

Area of Science:

• Neuroscience
• Linguistics
• Data Science

Background:

• Understanding neural processing of naturalistic speech is crucial.
• Existing datasets often lack detailed annotations or multimodal recordings.
• Bridging the gap between linguistic input and brain activity requires comprehensive data.

Purpose of the Study:

• To introduce the Kymata Soto Language Dataset, a novel resource for speech neuroscience.
• To provide synchronized electroencephalographic (EEG) and magnetoencephalographic (MEG) data with detailed linguistic annotations.
• To enable research on brain responses to natural conversational speech across different languages.

Main Methods:

• Collected EEG and MEG data from native Russian and English speakers listening to conversational speech.
• Recorded audio stimuli and created precise phoneme and word-level transcriptions with timestamps.
• Organized the dataset following the Brain Imaging Data Structure (BIDS) standards.
• Validated the data and preprocessing pipeline using Python-based analyses.

Main Results:

• The dataset contains multimodal neuroimaging recordings (EEG/MEG) and rich linguistic annotations.
• Analysis confirmed consistent low-level loudness perception trends across language groups.
• The BIDS organization ensures ease of use and compatibility with existing neuroimaging software.

Conclusions:

• The Kymata Soto Language Dataset is a valuable, open-source resource for studying the neural basis of naturalistic speech perception.
• The dataset's structure and comprehensive annotations facilitate reproducible research.
• Findings suggest shared low-level acoustic processing mechanisms across languages.