Project overview
The project aims to decode speech-related brain activity from non-invasive MEG recordings and reconstruct spoken and written language directly from neural signals. Experimental settings include listening, overt speech, silent articulation, and imagined speech in Spanish.
PhD position
The selected PhD candidate will lead the development of AI models capable of generating speech directly from MEG signals recorded during overt and imagined speech production.
Key Responsibilities:
• Develop brain-to-speech AI models from MEG data
• Progress from command classification to generative speech models
• Explore CNNs, transformer-based models, and diffusion architectures
• Contribute to MEG experiment design and dataset curation
• Implement evaluation protocols for intelligibility and generalization
• Release reproducible datasets and open-source pipelines
The project will explore transformer-based speech models such as Wav2Vec 2.0, HuBERT, and Whisper, as well as large language models including GPT and LLaMA variants.
Candidate Profile:
• MSc in AI, Computer Science, Biomedical Engineering, Computational Neuroscience, or a related field
• Strong background in machine learning and deep learning
• Solid Python skills (e.g., PyTorch or TensorFlow)
• Interest in speech processing and/or neural signal decoding
• Excellent written and spoken English skills
Experience with speech technologies or EEG/MEG is considered a plus.
Application Details:
Candidates should send their CV and a short motivation statement in English to: eva.navas@ehu.eus
• Expected start date: July 2026
• Funding: 4 years (full-time PhD contract)
• More information: brain2lang
All applications will be evaluated based on academic excellence, research experience, and alignment with the project goals.