Neural Noise

Real-time inner speech decoding with BCI for communication restoration

3 min read

  • A Stanford-led speech BCI has demonstrated decoding of inner speech (thoughts) in real time from intracranial signals, offering a pathway to restore communication for speech-impaired patients.1
  • The interface has been described as making the ‘inner voice’ audible for some patients and reflects BrainGate-style iEEG human trials.2
  • Real-time inner-speech decoding is implemented with AI and brain–computer interfaces for precision-medicine and clinical translation.3
  • Primary institutional and press coverage (Stanford Medicine, EurekAlert, Scientific American, NYT, Stanford University, Inside Precision Medicine) positions this as tier-1 for near-term translation.123456 12345

Gardner updates

  • Stanford-led speech BCI demonstrated decoding for communication restoration in speech-impaired patients. 1

  • Real-time inner-speech decoding from intracranial signals has been reported in human BCI trials. 2

  • High-profile coverage of inner-speech and speech prosthesis BCI reflects the same Stanford/BrainGate research thread. 3

  • Mainstream coverage highlights that for some patients the ‘inner voice’ may soon be audible via BCI. 4

  • Primary institutional summary describes a thought-decoding interface for the speech-impaired population. 5

  • Inner-speech decoding with AI and BCI is reported from a precision-medicine perspective. 6

  • An inner-speech decoder from neural data was reported; it raises mental privacy and BCI ethics issues—consent and policy for neural data are urgent. 7

Weekly enrichment (2026-07-20)

  • The underlying study is Kunz, Meschede-Krasa, Willett et al., “Inner speech in motor cortex and implications for speech neuroprostheses,” published in Cell on 14 August 2025 (188(17):4658–4673.e17; PMID 40816265).89
  • Investigators used multi-unit microelectrode array recordings from four participants with severe speech and motor impairment, finding that inner (imagined) speech is robustly represented in the motor cortex.810
  • Real-time decoding of imagined sentences achieved word error rates of 14–33% for a 50-word vocabulary and 26–54% for a 125,000-word vocabulary across participants T12, T15 and T16.810
  • In a proof-of-concept demonstration, the interface decoded imagined sentences from the ~125,000-word vocabulary with accuracy as high as 74%.11
  • Inner-speech neural patterns were highly correlated with attempted-speech patterns but separable along a distinct “motor-intent” dimension, enabling systems to distinguish the two.810
  • A password-gated safeguard (users imagined “chitty chitty bang bang”) was recognized with more than 98% accuracy and prevented the BCI from decoding private inner speech unless intentionally unlocked.11
  • Participants preferred inner speech to attempted speech mainly because it required less physical effort and was less fatiguing.812
  • The work was a BrainGate-style collaboration spanning Stanford, Emory University, Georgia Institute of Technology, UC Davis, Brown University and Harvard Medical School.12
  • Some free-form private inner speech could be decoded during counting and sequence-recall tasks, underscoring mental-privacy and consent concerns for neural-data policy.810

Footnotes

  1. https://news.google.com/rss/articles/CBMiggFBVV95cUxPSHpXRkljdUFCdE1HT05qLWZYZzdJX2VLcXZCeHVvOWQtbm5qTEt0ZVBtcS12cVJSbkUxUmpobnYxUzlrQVdibVlOS2pONml1UHV5cmxhbTVNR0Jjd2s4NHBUSDdmN0hudnBNU0RJVmV0WFAtU0xEQkpkV3dGY1dFeVlB?oc=5 ↩ ↩2 ↩3 ↩4

  2. https://news.google.com/rss/articles/CBMiXEFVX3lxTE1BdmpDOV8xM0p6MmlGV3lJYjh5MmVlTTlHZFdIMG5LRlk4Wkt3QTlRWVl5WUtFT3VZbGRJd1FGVVdIRkM4RzFPbVhpcnlkTnAyZ1BpT2l5NGJKMVRX?oc=5 ↩ ↩2 ↩3 ↩4

  3. https://news.google.com/rss/articles/CBMimgFBVV95cUxPUEJnM1pZRUZxeXoyVENMN2Njc1VZdzd3NUJxR20tT054ZFFwVzM5LWVmSV9zT1VGNk5zY3BkcFZ1M01XTi0tbFEwZjhHRDJPVURmOXgyU2Frc1VaSXppUHNHQ2FqNW5hR0NPYnBOVTBDVnZWbDdWc2NmUGxpVDFtRWk3a3BOdF9vNzU5Sm53V1U1OHBMd0hIeDZB?oc=5 ↩ ↩2 ↩3 ↩4

  4. https://news.google.com/rss/articles/CBMiiwFBVV95cUxQTzJCNU4xcUhOQVJEWE5zWWJRc3RxM0ZjWXVYTXBudkI4UkI1UTRxcFVmRjNfTm1IbjIzNVl2VEl3ODF4Z2FHNmpQQTFEeDdvTDRqazJVWTBQMER5SEdGd05ZaDRCSUpEcnhXUVNNUzlRMnRlNWF2eklNN1lGanI3VTJiemhqWTVCbFhR?oc=5 ↩ ↩2 ↩3

  5. https://news.google.com/rss/articles/CBMimwFBVV95cUxOMDFnNS1XbXg0T0t3Rzh0TzNTY2xnRFNDLW96VkhwQzhzckllWldYSXBScXlZQUpHN3JwblpvT3dkZEgtWFVNQVkyZnJaaUNHZlpJQl8wWW5uRUc1NVZxMEs3b0hsWVNkR3VhMnludGRlQnhzN3VEY1NZV29MLW83RU9TaTA2Mjk2ZmdNMW1Kd1ZLV1d2ZlhaeUd5cw?oc=5 ↩ ↩2 ↩3

  6. https://news.google.com/rss/articles/CBMizAFBVV95cUxOa0dFU2ViVUdVUjdGWlh6QUpkNWNxdUdDZzRTcVJxWXpUeDg1VEVCZmNyZXFISFZvYnFNOGVRUWExSWp0Z1hvRFNBeldhLUpndzFNUzU1NDQwUWNJQWNTTEVhVWlfM0dWSlNsZTFuVnUzLW5XQWZ4UlJTNDBSaW15S25GakN2RzBJZTRYOUNnM1VxRnFybmNjM09WT0hIY2xqVndMRHNia1Zsbm9hOExIY25ZMEg1djRVQldoR0JScFp5RlBXYS1HMGlWLXo?oc=5 ↩ ↩2

  7. https://news.google.com/rss/articles/CBMioAFBVV95cUxNOHhGaHU0UzIxOGhyTzdtc1F2ODJISEp5OURaV0dWVzN0bmtaWmQ1Y2U3dVpzYWdONGRJd3p0M2xsdWd2MXdQOUNRSlZ6eFFZQjNBbWVfdU9zNEpzdVlTRlJzcUNSNEtPdmM2OHdWUzNYNWszcXpkUVczTWtCVXZMRV9YSExiYXRPX1QzMzNfaTJjSTVEV01kUWpUVkdfM2cx?oc=5 ↩

  8. https://www.cell.com/cell/fulltext/S0092-8674(25)00681-6 ↩ ↩2 ↩3 ↩4 ↩5 ↩6

  9. https://europepmc.org/article/med/40816265 ↩

  10. https://www.nih.gov/news-events/nih-research-matters/decoding-inner-speech-brain-signals ↩ ↩2 ↩3 ↩4

  11. https://www.eurekalert.org/news-releases/1093888 ↩ ↩2

  12. https://news.stanford.edu/stories/2025/08/study-inner-speech-decoding-device-patients-paralysis ↩ ↩2

Graph View