spec2wav
Experimenting with a neural vocoder derived from other researches The skeleton is the same as SampleRNN, with modifications in architecture to suit my use case.
Will modify the architecture to make it extremely flexible for multiple purposes. Speech, Music, Any Other Audios, etc.
TODO
Training Block
- Saving Model
- Loading Model
- Evaluation?
- Train/Val loss visualisation
- Generation audio every epoch
Model
- Model Input Output Tuning
- Weight/Hidden State Initialization
- Weight Normalization
- Generator File/ Module
- HyperParameter Tuning
- CUDA
- SalesForce QRNN
- Handle Variable Length Input
- Gaussian Noise?