seannobel / speech-decoding Goto Github PK

Hi, thank you for your implementation! I wonder that if you get the same accuracy on gwilliams2022 dataset as the paper said? I'm trying to re-train the model, but it is overfitting all the time! It really confuses me because I find that I have the same preprocess pipeline as the paper(and as yours).
And by the way, have you read the author's newest paper BRAIN DECODING: TOWARD REAL-TIME RECONSTRUCTION OF VISUAL PERCEPTION ? I failed to re-train that model either. :(

Hi. Unable to access shared link for the merged version of the audio files.

Hydra is:

1. Based on YAML config files (with arguments that can be overwritten on the command line)
2. Will let you compose configs so some hyperparameters can be shared across datasets
3. Automatically creates timestamped folders for experiment runs
4. Compatible with W&B

Would replace configs/args.py (with some config moved into a "base" YAML config file) and configs/parser.py. No need to write experiment names or write into the runs/ folder - Hydra will automatically make an output folder and put experiments in separate folders there.