Neural Noise- EEG oscillations and related brain generators for phonation phases in long utterances inform iEEG/EEG-based speech prostheses and neural signal processing (Nature).1
- The work directly supports speech BCI and neural decoding with electrophysiology focus and high relevance to decoder design.1 1
Weekly enrichment (2026-07-20)
- The source study (Scientific Reports, 2025) recorded 64-channel scalp EEG in 19 participants performing a prolonged-phonation paradigm (after Ladefoged 1967): continuously repeating the syllable “pa” on a single breath for as long as possible, with the utterance divided into four phonation phases.2
- Neural generators accounting for the reactivities were estimated with a distributed inverse solution (swLORETA), tying each phase transition to identifiable oscillatory activity.2
- Across the four phases, two sustained bands of event-related desynchronization (ERD) appeared — delta (1–3.5 Hz) and high alpha (10.5–13 Hz) — alongside a more diffuse event-related synchronization (ERS) spanning theta (4–7 Hz) to low alpha (7.5–10 Hz), plus scattered short beta (15–20 Hz) clusters.2
- Estimated generators spanned frontal regions at vocal initiation and, across later phases, premotor/motor cortex, anterior cingulate, sensorimotor areas, thalamus, and cerebellum, consistent with coordinated respiratory-articulatory control.2
- The authors connect these scalp findings to intracranial evidence that the breathing cycle modulates iEEG gamma-band envelope oscillations, with volitional-breathing networks reported in caudal-medial frontal, premotor, orbitofrontal, and motor cortex, plus insula, superior temporal gyrus, and amygdala.2
- Intracranial (ECoG) work shows speech production is marked by high-gamma augmentation (~70–200 Hz) together with a beta-band (15–25 Hz) power decrease over motor and temporal cortex — the exact signal features that speech BCIs typically decode.3
- In a landmark ECoG study of all 44 English phonemes, gamma augmentation in the larynx-sensorimotor area occurred uniformly at sound onset and reliably distinguished voiced from voiceless phonemes, supporting phonation-phase decoding.3
- Laryngeal motor cortex also encodes finer phonatory features: lexical tones were decodable from LMC neural populations as early as 300 ms before pitch onset, peaking at 66.7% accuracy, relevant to pitch and phonation control in speech neuroprostheses.4
- A practical caveat for iEEG speech BCIs: speech itself produces mechanical vibration artifacts that create a narrowband gamma component tracking the voice fundamental frequency, which can mimic physiological high gamma and must be identified and filtered.5