This repository handles the training for my science fair project, predicting ragas from an audio input file. The total project includes 5000+ lines of code across different branches and iterations of the system.
File | Description |
---|---|
train.py | Actually manages backpropagation, network states, and access data loaders. Reports all training and validation accuracies to a TenssorBoard data logger. |
data.py | Handles data loading during training. |
models.py | Network model files, it's a class for models |
utils.py | Various training utilities |
Branch | Description |
---|---|
Master (Stage #1 Training) | Stage #1 training of simple convolutional network on chunks of data |
Recurrent | Conducts Stage #2 of training with additional two-layer LSTM network. |
Leave One Validation | Conducts thorough leave-one validation benchmark comparison for full recurrent PhonoNet architecture |
If you have any questions or need assistance with this repository in any way, do not hesitate to contact me at sauhaarda@gmail.com. Personal website available here as well.