A multimodal speech-production dataset with time-aligned articulography, EEG, audio, and vocal-tract anatomy
2026-07-02
We present a multimodal speech-production dataset combining simultaneous electromagnetic articulography (EMA), electroencephalography (EEG), and audio from 29 adult native speakers of German. All participants have external craniofacial anthropometry; an anatomical subset (N = 18) additionally contributed acoustic pharyngometry, rhinometry, and 3D head surface meshes. Speech materials include high-trial diadochokinetic sequences at habitual and maximally fast rates, plus EMA+audio for passage reading and sustained vowels, as well as EMA for palate tracing and non-speech oromotor actions. The corpus contains 8,700 syllable-task trials and approximately 17 h of EMA with matched audio. A microcontroller mirrors each EMA sweep’s start and stop as 1 ms transistor—transistor logic (TTL) pulses on the EEG digital channel, enabling sub-millisecond alignment. We distribute raw and minimally processed streams, stable event codes, and machine-readable metadata, plus example Python utilities (and a container) for loading, synchronising, and basic preprocessing. The resource supports studies that exploit articulatory landmarks for EEG alignment, examine pre-movement activity, assess overt-speech EEG artefact handling, and develop anatomy-informed models linking vocal-tract structure to articulatory dynamics and acoustics. Data and code are openly available under a CC-BY 4.0 licence with versioned DOIs.