CIPHER: Conformer-based Inference of Phonemes from High-density EEG
📰 ArXiv cs.AI
CIPHER is a dual-pathway model for inferring phonemes from high-density EEG using conformer-based architectures
Action Steps
- Extract ERP features and broadband DDA coefficients from high-density EEG signals
- Implement a dual-pathway conformer-based model to integrate the extracted features
- Train the model on a dataset with concurrent TMS, such as OpenNeuro ds006104
- Evaluate the model's performance on binary articulatory tasks and assess its vulnerability to confounds
Who Needs to Know This
Neuroscience and AI researchers can benefit from this model to improve speech decoding from EEG signals, and software engineers can implement and fine-tune the model for real-world applications
Key Insight
💡 CIPHER achieves near-ceiling performance on binary articulatory tasks, but is highly vulnerable to confounds
Share This
💡 CIPHER: Conformer-based Inference of Phonemes from High-density EEG Representations
DeepCamp AI