Brain2voice 2.0: High-performance voice synthesis brain-computer interface

bioRxiv · 13 Nov 2026 · Preprint · Open access

Wairagkar, Srinivasan + 8 more

The abstract is read at the publisher; the summary is JClub's.

Brain2voice 2.0 synthesizes highly intelligible real-time voice from intracortical neural signals, achieving 5.24% word error rate—an 8x intelligibility improvement over prior SOTA.

Why it matters: Existing BCIs decode speech as text but cannot provide instantaneous voice, and current brain-to-voice output remains too unintelligible for real-world use.
What they did: The multimodal Transformer decoder is trained on continuous, custom-tokenized acoustic targets plus phoneme targets using self-supervised and adversarial objectives, producing outputs every 10 ms.
The result: On the prior intracortical brain-to-voice benchmark dataset, listeners achieved 5.24% word error rate (8x better than 43.75% prior SOTA), with raw phoneme error rate of 7%.

AI summary of the abstract · may contain errors · not medical advice · read the paper.

Open access

Free. We’ll email you when a new paper cites this one or any other paper you follow on JClub — at most one email a day. Every email has a one-tap unsubscribe.

Geometry-Constrained Coupling Drives Critical Dynamics Across Neural Manifolds
bioRxiv · 13 Nov 2026 · Preprint
AI summary · may contain errors
Geometry-constrained coupling between cortical and hippocampal neural fields yields distinct dynamical regimes, preserving intrinsic alpha (8–12 Hz) and theta (4–8 Hz) rhythms at moderate coupling.
- Open access
- 3 cites
Open
Share
Post on X Post on Bluesky Share on LinkedIn
https://jclub.com.au/a/10.1101/2025.09.07.674721
Open in JClub
Closed-loop neurofeedback reshapes preparatory brain states to bias subsequent pain processing
bioRxiv · 13 Nov 2026 · Preprint
AI summary · may contain errors
Authentic EEG-contingent closed-loop neurofeedback in a double-blind sham-controlled 2×2 design enhances targeted somatosensory cortex preparatory power and reduces pain intensity.
- Open access
Open
Share
Post on X Post on Bluesky Share on LinkedIn
https://jclub.com.au/a/10.64898/2026.06.02.729491
Open in JClub
Restoring intended speech in people with dysarthria from pontine stroke using an intracortical brain-computer interface
medRxiv · 10 Nov 2026 · Preprint
AI summary · may contain errors
Intracortical brain-computer interfaces can decode intended speech in people with pontine stroke–related dysarthria using as few as one 64-channel microelectrode array, reaching 19.4% WER (125,000 words) for T16.
- Open access
- 6 cites
Open
Share
Post on X Post on Bluesky Share on LinkedIn
https://jclub.com.au/a/10.64898/2026.02.19.26346583
Open in JClub

Journals this month

All journals →

Topics moving, week to 14 Nov 2026

All topics →

By field

All journals by field →

Brain2voice 2.0: High-performance voice synthesis brain-computer interface

Related in JClub

Geometry-Constrained Coupling Drives Critical Dynamics Across Neural Manifolds

Closed-loop neurofeedback reshapes preparatory brain states to bias subsequent pain processing

Restoring intended speech in people with dysarthria from pontine stroke using an intracortical brain-computer interface

Browse JClub