speech-decoding

Subvocal 推出了一款可通过面部雷达感知无声口型、以 200+ WPM 将默读词语转为文字的智能眼镜，实现无键盘、无语音的输入方式，并已获 YC 与 Afore Capital 的 130 万美元融资。

该论文发现非侵入式脑电解码语音中报告的大幅改进很大程度上可由时间间隔捷径在无脑数据的合成信号上复现；通过将句子内的滑动窗口独立编码以移除该捷径后，提出的SimpleB2T方法在五次观测下达到36.6%的词错误率，显著提升了脑到文本解码性能。

This paper proposes a multi-subject pretraining approach for surface EMG speech decoding that reduces calibration time while improving accuracy, achieving a character error rate of 21.7% with just three minutes of target-subject data.

Neuralink's brain-computer interface enables an ALS patient to regain speech by decoding neural signals in real-time, demonstrating a significant breakthrough in communication technology.

Myovox decodes open-vocabulary English text from 31-channel surface electromyography recorded from facial muscles, achieving an 18.53% word error rate on the emg2speech General Corpus through decode correction, bidirectional Conformer training with cross-modal distillation, and language model reranking.

This paper introduces Brain2Semantics2Text, a non-invasive speech decoding method that maps MEG responses to semantic embeddings to reconstruct sentence-level text without word-level alignment.

The article announces the 2026 PNPL Competition focused on word classification and efficient cross-subject generalization in the LibriBrain100 dataset, advancing non-invasive speech decoding for brain-computer interfaces.

The paper introduces LibriBrain100, a large-scale MEG dataset with over 100 hours of data for neural speech decoding, achieving state-of-the-art performance on word classification benchmarks.

This research uses non-invasive EEG and contrastive learning to decode words during silent reading, showing scalable lexical information recovery that scales with data volume.

This paper presents an interpretable MEG-to-audio retrieval model for perceived speech, redesigned with spherical-harmonics spatial attention and source mapping, achieving 39.75% Top-1 accuracy with far fewer decoder parameters while revealing which speech features drive retrieval.

This paper investigates the effects of phonetic versus character targets and selective state-space models (Mamba) for intracortical brain-to-text decoding, finding that a GRU-based recurrent decoder remains the strongest performer on the Brain-to-Text '25 benchmark.

A model trained on ultrasound recordings of the tongue can decode silent speech with a 15.6% word error rate, approaching lip-reading performance despite a smaller dataset.

A new non-invasive method translates brain waves into words, offering a communication pathway without the need for surgery.

Casey Harrell, a man with ALS, has become the first 'power user' of a speech brain-computer interface, using it independently for thousands of hours to speak, browse the web, and work. A study in Nature Medicine reports the device's long-term reliability and expanded capabilities.