Papers

A study proposes a hierarchical alignment framework to retrieve a read sentence using EEG signals and context. It focuses on sentence-level EEG-to-text classification, rather than directly generating sentences or identifying arbitrary thoughts from EEG alone. The authors note that much prior decoding research has focused on phonemes, letters, and words, while sentences carry richer meaning. They also describe sentence classification as less studied than sentence generation. The framework explores EEG-text relationships to inform the choice between retrieval and generation. Non-invasive assistive communication is its application context, not an established benefit for people with communication disabilities.

University of Melbourne researchers studied one participant implanted with a Stentrode endovascular brain-computer interface who attempted to handwrite English characters. They classified the attempts offline and online using neural signal trajectories over time. Across four online sessions, each containing 20-30 trials per character and 80-180 rest trials, the best-performing 3-class model achieved a mean rest recall of 87%, character recall of 65%, and character precision of 59%. Adjusting all trials to the same duration significantly improved classification compared with raw trials. The findings support further investigation of temporal information for motor decoding, but come from one participant and a small number of classes.

A prospective single-centre study involving Columbia University Medical Center found that 13 of 41 clinically unresponsive, critically ill patients with acute brain injury (32%) could control a brain-computer interface. The system analyzed EEG in real time and translated attempted movement into changes in sound. Clinical consciousness returned by discharge in 4/13 patients with BCI control versus 1/28 without it. At 12 months, 6/13 versus 2/28 met the study's definition of functional independence. BCI control was associated with subsequent recovery, but the findings do not show that device use caused recovery. The tested interface was one-way; bidirectional communication remains a future possibility.

A University at Buffalo author evaluated an assistive communication interface combining a BCI with eye tracking. Gaze narrows the target area and brain signals confirm selection. The study reports faster selection than BCI alone, better control than eye tracking alone, and fewer selection errors and corrections.

University of Essex researchers recorded EEG from 14 participants retrieving concepts from verbal clues. Mean accuracies in three pairwise comparisons of food, musical instruments and body parts were 73.4%, 70.7% and 68.2%. The model learned representations from other participants before calibration with each test participant’s own labelled data.

Researchers affiliated with Araya Inc. and Sincxpress Education SRL examined a model that predicts spoken words from ultra-high-density EEG to assess whether decoding reflects neural activity rather than electrical activity from speaking muscles. Electrode contributions could not be explained by their mutual information with EMG, and the periods contributing to EEG decoding differed from those supporting EMG-only decoding. Contributing periods partially overlapped between overt and minimally overt speech. Denoising increased contributions from electrodes in speech-related brain areas across all speech conditions. The authors interpret these findings as supporting neural speech decoding with appropriately preprocessed EEG.

Researchers at the University of Nebraska–Lincoln and University of Nebraska at Omaha compared three communication displays in 24 neurotypical children aged 8–13. Scene and grid displays had similar offline BCI performance, but children rated scenes as more visually appealing. Layout and background also affected EEG responses.

Oxford researchers compared six aggregation strategies; training-time logit averaging led in six of seven settings. Aggregating 100 same-phoneme trials yielded 64.1% balanced accuracy in a public-test-split re-evaluation, 17.4 points above the competition champion model. The preprint experiments use MEG recordings.

Researchers proposed BEST, a two-stage method that converts neural activity into text, to examine whether speech brain-computer interfaces need large models. On the Brain-to-Text '24 and '25 benchmarks, BEST achieved word error rates of 6.19% and 1.62%, respectively, using only 3.85% of the parameters required by comparable methods. Word error rate measures errors relative to reference text; lower values indicate fewer errors. The authors argue that larger decoders do not consistently perform better and that alignment between neural signals, speech, and text matters more than model size. These benchmark results do not directly establish everyday communication performance.

Researchers audited three public corpora and found that decoding overt speech may not establish a pipeline's ability to recognize brain-derived speech signals. In ds003626, two lip electrodes achieved an accuracy of 0.7430, compared with 0.4231 for the 128-channel scalp montage; no decodable signal remained in the 8-30 Hz band. By injecting effects of known amplitude into real recordings, the team also measured detection floors and found differences between methods. The authors conclude that muscle signals dominate the overt-speech positive controls in these corpora and recommend reporting band-limited performance, artifact-channel accuracy, and measured detection floors.

Researchers analyzed three public corpora to test whether imagined speech decoders trained on one person's scalp EEG could work for another. Across 482 ordered participant pairs, none remained significant after multiple-comparison correction; mean transfer accuracy was within 0.006 of chance in every corpus. A signal distinguishing the imagery task from pre-stimulus rest instead reflected a post-stimulus state. Only the competition corpus exceeded the within-subject threshold for every participant, but its frequency profile failed an artifact audit. The findings apply to these corpora and the representations examined, not to all imagined speech decoding.

Researchers aligned EEG recorded during imagined speech with text representations to exploit semantic relationships beyond fixed intent labels. Evaluated on five subjects and 39 intent categories in the Chisco dataset, the framework performed comparably to a task-optimized closed-set classifier. Maximum EEG-text similarity identified held-out trial subsets with higher Top-1 and Top-5 accuracy. Mean agreement exceeded the 50% chance reference on all five semantic dimensions. Comparisons with unseen candidate sentences were qualitative demonstrations, not evidence of unrestricted imagined-sentence decoding.

Researchers developed a non-invasive EEG brain-computer interface combining auditory input with visual feedback to support phonological processing relevant to developmental dyslexia, then tested it first in adults with and without dyslexia. Participants completed 16 sessions and underwent pre- and post-training assessments of phonological awareness, single-word reading and non-word reading. Among participants who learned the BCI, offline learning scores were significantly associated with non-word reading gains in both groups. BCI learning was also significantly associated with improvements in real-word reading and amplitude rise time discrimination. These associations do not establish that training caused the gains or demonstrate benefits in infants or children.

Researchers used natural-language class descriptions to help models learn the correspondence between brain signals and target classes, evaluating the approach in visual, motor, and speech decoding. The study tested seven encoder architectures with independently trained models using scalp and intracranial recordings. Speech-decoding accuracy improved by up to 17.15 percentage points over matched class-label-supervision baselines. The learned representations also supported candidate-set reuse and label-supervised subject and session adaptation. In steady-state visual evoked potential tasks, the method enabled unseen-frequency decoding under specific conditions. These findings do not establish a single model that performs every task or adapts to new subjects without labels. This study is a preprint and has not been peer reviewed.

A contextual prior can raise overall accuracy in MEG speech decoding while making some correctable errors more confident, a preprint reports. Tracking candidate scores across two magnetoencephalography datasets, the authors describe a reversal: when the correct candidate starts near the top of the local ranking, a larger fused margin makes a repair more likely, but when it starts lower, a larger margin makes the error more likely to persist. On MEG-MASC, pooled correctness AUROC is 0.87, while the AUROC separating repairs from residual errors falls from 0.70 to 0.39; errors starting beyond rank 20 account for 46.6% of post-fusion errors. The authors argue confidence should retain local and contextual evidence rather than fused scores alone; reading the two scores separately lifts decoder coverage to 74.5% of windows from 56.7%. The study is a preprint and has not been peer reviewed.

Researchers recorded EEG during overt and imagined vocalisation to distinguish five emotion categories: anger, happiness, neutral, sadness, and pleasure. The overt experiment involved 14 participants and the imagined experiment involved 21, yielding single-trial decoding accuracies of 78.1% and 36.4%, respectively, against 20% chance. Facial surface EMG recorded in five participants during imagined vocalisation indicated that involuntary muscle activity contributed to EEG spectral and spatial patterns used for classification. One lateral-channel pattern was stronger during imagined happy vocalisations. The results mean that decoding accuracy cannot be attributed entirely to brain-derived emotion information.

An ISAE-SUPAERO study tested a brain-computer interface combining semi-dry EEG acquisition with visual stimulation. Thirty participants completed an 11-class keypad typing task on a standard screen. With 33 s of calibration data, the system achieved 99.8% accuracy and a mean decoding time of 4.3 s. No spurious commands occurred during a 2-min video-viewing idle condition between two online typing sessions. A virtual reality proof of concept reached 100% online accuracy with a mean decoding time of 5.6 s. Participants reported high visual comfort and little peripheral distraction in both conditions, but greater fatigue in VR. The short idle test does not establish long-term performance.

Researchers have proposed NEUROTOKEN, a single network that uses conditional flow matching to score candidate voices and identify, from EEG, which speaker a listener is attending to in a noisy room, the so-called cocktail-party problem. On the public KU Leuven, DTU and NJU datasets with 5-second windows, it raised per-segment accuracy by 9% to 16% over the strongest non-generative baseline and narrowed across-subject variance by about 3x. The authors also report that the 95% to 97% accuracies often cited for decoding attention direction fall by 17% to 45% once training and test trials are strictly separated. The work is a preprint and has not been peer reviewed.

Speech brain-computer interfaces aim to restore communication by turning neural activity tied to speech, language or communicative intent into text, synthesized voice or avatar movements. This review argues they are not simple neural-to-text decoders but clinical systems that change over time, in which neural representations, recording hardware, decoding models, language models, feedback and the user's own learning interact. The authors say the next generation should be judged not only on offline accuracy but on stability across sessions, calibration burden, latency, the ability to signal uncertainty, usability and safeguards against decoding words a user did not mean to say. The paper is a preprint and has not been peer reviewed.

Gabriele Cinà reported in an arXiv preprint posted on September 27, 2026, that Jev, a hosted model, can replace the multi-billion-parameter language model that speech neuroprostheses use to rescore candidate sentences, the only GPU-dependent step in the system. Jev returns a probability for each candidate in one call, which is combined with the decoder's own score. On 978 held-out sentences from a participant with ALS, the published decoder alone had an 8.1% word error rate, versus 7.5% with Jev and 7.8% with both OPT-6.7b and Qwen2.5-7B; after re-tuning decoder weights, the figures were 6.9%, 7.2% and 7.4% respectively. Jev needs no GPU, costs $0.07 per thousand sentences, and has 262 ms end-to-end latency over the internet, versus 27 ms for a 7B model on a local GPU. The work is a preprint and has not been peer reviewed.

Researchers have proposed SICMD, a method that combines two non-invasive brain signals, functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG), to identify speech a person is hearing from brain activity with a single model that works across different people. fMRI pinpoints where activity occurs and MEG captures when, together giving fuller spatiotemporal information. In cross-subject perceived-speech decoding, the method improved Top-1, Top-10 and Rankacc by more than 10.6%, 10.1% and 1.7% over baseline methods, and cut training costs by 88.8% and 60.5% compared with multi-subject and intra-subject decoding respectively. The work is a preprint and has not been peer reviewed.

Researchers have shown that language models do more to help decode imagined speech than heard speech from non-invasive brain recordings. Imagined speech is harder to read than speech a person hears: the brain's response is weaker and noisier, and less data is available. Using a magnetoencephalography (MEG) dataset in which participants both listened to and imagined the same material, the team brought a language model into both training and decoding. Imagined speech gained more from the model than heard speech, and its best results required leaning more heavily on the model. The authors conclude that the weaker the neural evidence, the more useful a language model's contextual prior becomes. The study is a preprint and has not been peer reviewed.

Researchers re-examined a multi-class imagined speech EEG dataset, words spoken silently in the mind, under a strictly subject-independent protocol in which a model is never trained on data from the person it is tested on. Spectral bandpower features reached a mean trial-wise accuracy of 49.03% ± 4.18% in coarse-level classification, clearly ahead of the 37.97% ± 3.79% from time-domain statistical features, and a few frequency bands carried most of the discriminative information. The study is a preprint and has not been peer reviewed.

Wearable neurotechnology and BCI applications, from assistive interfaces to clinical monitoring, favor single-channel EEG because it needs few electrodes and is light to wear. But muscle, eye and motion artifacts are hard to remove from a single channel, especially without auxiliary channels, artifact labels or hand-picked clean baseline segments. SPAR-EEG needs none of these: it applies three artifact-specific attenuation passes to each EEG segment, one based on variational mode decomposition for high-frequency muscle bursts and two based on singular spectrum analysis for blink-like ocular transients and slow motion drift. It achieved the best artifact-region SNR improvement at all 26 SNR levels tested and raised final accuracy by 7.8 percentage points in a P300 speller task using only the FP1 and FP2 leads.

Researchers developed a non-invasive EEG brain-computer interface using auditory inputs and visual feedback to target phonological processing in developmental dyslexia, initially testing it in adults with and without dyslexia. Participants completed 16 sessions and underwent phonological awareness and reading assessments before and after training. Among participants who learned the BCI, including both dyslexic and control participants, offline BCI scores were significantly associated with improvements in nonword reading. BCI learning also showed significant associations with gains in real-word reading and amplitude rise time discrimination. These associations do not establish causation or demonstrate benefits in infants or children. This study is a preprint and has not been peer reviewed.

P300 BCI spellers are often built by switching on every component that works on its own, on the assumption that more is better. A preprint study tested that assumption with a four-component full-factorial experiment and found that a component's value is conditional, not additive. Subject calibration was the strongest single contributor; Euclidean Alignment made up for the lack of calibration in zero-calibration settings; and stacking components that are each useful on their own could lower performance, an effect the authors call component anti-synergy. Language model support was not universally beneficial either: its effect depended strongly on how strong the underlying EEG pipeline was.