Projects
A slightly mad lab for ideas that venture out without a guide.
Filter By
ASR for dysarthria
Real-time speech recognition adapted to dysarthric voices. Two-stage pipeline: fine-tuning wav2vec2 on healthy speech, then speaker adaptation via LoRA to avoid catastrophic forgetting. My ENSEIRB-MATMECA capstone, driven by accessibility (+28 accuracy points over zero-shot on the target speaker).
PythonPyTorchwav2vec2LoRA / PEFTHugging FaceASR
PoseNet
A research project on action recognition using human pose data.
PythonPose EstimationOpenCVDeep Learning
WitHub
A comprehensive toolkit to automate project versioning on Git using AI.
PythonGitOllamaLLMs
