back/README.md

# Inversynth Fork
## AMP Team 

## Launch instructions :

*Optional: outputing your own config file for your VST* 
```zsh
python -m generators.vst_generator generate
```
*1. Dataset Creation based on config profile* 
```zsh
python -m generators.vst_generator run --config "your_config_path.json"
```

*2. Model training* 
```zsh
python -m generators.spectrogram_cnn --epoch "your_epoch_number" --model C6XL
```

Parameter | Default | Description
---|---|---
`--num_examples` | `2000` | Number of examples to create
`--name` | `InverSynth` | Naming convention for datasets
`--dataset_directory` | `test_datasets` | Directory for datasets
`--wavefile_directory` | `test_waves` | Directory to for wave files.<br>Naming convention applied automatically
`--length` | `1.0` | Length of each sample in seconds
`--sample_rate` | `16384` | Sample rate (Samples/second)
`--sampling_method` | `random` | Method to use for generating examples.<br>Currently only random, but may<br>include whole space later
Optional |
`--regenerate_samples` | | Regenerate the set of points to explore if it<br>exists (will also force regenerating audio)
`--regenerate_audio` | | Regenerate audio files if they exist
`--normalise` | | Apply audio normalization

This module generates a dataset attempting to recreate the dataset generation<br>as defined in the [paper](paper/1812.06349.pdf)

Selecting an architecture:

- `C1`, `C2`, `C3`, `C4`, `C5`, `C6`, `C6XL`