Project based of this template: https://github.com/victoresque/pytorch-template#pytorch-template-project
- Connect to fix-it-in-post.net repo
- Remember to use git credentials and access keys
- Create configuration to run each time notebook is start to install tensorboard on startup, using
./sagemaker_scripts/setup.sh
Processes the files. Passing all will process all the data, by default it will process 512 samples
python process.py [all]
The following command will run the training script. If the data exists locally or on S3, the data will be pulled from those locations. If not, the data will be processed locally instead.
python train.py --config config/baseline_fc.json
If you have data processed locally, it would be good to store on S3 for others to use without having to reprocess the data. You can store the data on S3 as such:
python upload.py ./data/processed/edinburgh-noisy-speech-db/w256o75sr8000n8/
This will store the data under the key w256o75sr8000n8/sample.0.pkl in the bucket fix-it-in-post
We use synthesis.ipynb to check performance of certain model - by listening to clean, noisy ans denoised samples. It is also possible to calculate SNR and PESQ for certain sample and also in aggregated manner.