iPhoneme Brain-to-Text System Aims to Restore Speech for ALS Patients Using ConformerXL Decoding

There's this new brain-computer interface called iPhoneme that's been developed to help people with ALS-related dysarthria, which affects between 173,000 and 232,500 individuals worldwide. It tackles two big issues that previous speech BCIs struggled with: how accurately they decode neural signals and how easy they are to use. The system uses a deep learning phoneme decoder based on a special Conformer architecture with about 192.9 million parameters, and it has a gaze-assisted input that reduces problems common in eye-tracking. The acoustic model features advanced techniques to stabilize communication, but so far, only 22 to 31 patients have been tested due to these challenges. This research can be found on arXiv under the identifier 2604.16441v1.