Merge pull request #4 from Stability-AI/os_release_2

Rename to stable-audio-tools and remove unsupported features for open-source release

Author: zqevans

README.md

@@ -1,11 +1,11 @@
-# harmonai-tools
+# stable-audio-tools
 Training and inference code for audio generation models
 
 # Install
 
 The library can be installed from PyPI with:
 ```bash
-$ pip install harmonai-tools
+$ pip install stable-audio-tools
 ```
 
 To run the training scripts or inference code, you'll want to clone this repository, navigate to the root, and run:
@@ -37,7 +37,7 @@ $ python3 ./train.py --dataset-config /path/to/dataset/config --model-config /pa
 The `--name` parameter will set the project name for your Weights and Biases run.
 
 ## Training wrappers and model unwrapping
-`harmonai-tools` uses PyTorch Lightning to facilitate multi-GPU and multi-node training. 
+`stable-audio-tools` uses PyTorch Lightning to facilitate multi-GPU and multi-node training. 
 
 When a model is being trained, it is wrapped in a "training wrapper", which is a `pl.LightningModule` that contains all of the relevant objects needed only for training. That includes things like discriminators for autoencoders, EMA copies of models, and all of the optimizer states.
 
@@ -97,7 +97,7 @@ Additional optional flags for `train.py` include:
   - RNG seed for PyTorch, helps with deterministic training
 
 # Configurations
-Training and inference code for `harmonai-tools` is based around JSON configuration files that define model hyperparameters, training settings, and information about your training dataset.
+Training and inference code for `stable-audio-tools` is based around JSON configuration files that define model hyperparameters, training settings, and information about your training dataset.
 
 ## Model config
 The model config file defines all of the information needed to load a model for training or inference. It also contains the training configuration needed to fine-tune a model or train from scratch.
@@ -118,7 +118,7 @@ The following properties are defined in the top level of the model configuration
   - The training configuration for the model, varies based on `model_type`. Provides parameters for training as well as demos.
 
 ## Dataset config
-`harmonai-tools` currently supports two kinds of data sources: local directories of audio files, and WebDataset datasets stored in Amazon S3.
+`stable-audio-tools` currently supports two kinds of data sources: local directories of audio files, and WebDataset datasets stored in Amazon S3.
 
 # Todo
 - [ ] Add documentation for dataset configs

defaults.ini

@@ -2,7 +2,7 @@
 [DEFAULTS]
 
 #name of the run
-name = harmonai_tools
+name = stable_audio_tools
 
 # the batch size
 batch_size = 8 

docs/autoencoders.md

@@ -7,7 +7,7 @@ The *decoder* takes in a d-channel latent sequence and upsamples it back to the
 
 Autoencoders are trained with a combination of reconstruction and adversarial losses in order to create a compact and invertible representation of raw audio data that allows downstream models to work in a data-compressed "latent space", with various desirable and controllable properties such as reduced sequence length, noise resistance, and discretization.
 
-The autoencoder architectures defined in `harmonai-tools` are largely fully-convolutional, which allows autoencoders trained on small lengths to be applied to arbitrary-length sequences. For example, an autoencoder trained on 1-second samples could be used to encode 45-second inputs to a latent diffusion model.
+The autoencoder architectures defined in `stable-audio-tools` are largely fully-convolutional, which allows autoencoders trained on small lengths to be applied to arbitrary-length sequences. For example, an autoencoder trained on 1-second samples could be used to encode 45-second inputs to a latent diffusion model.
 
 # Model configs
 The model config file for an autoencoder should set the `model_type` to `autoencoder`, and the `model` object should have the following properties:

docs/pretransforms.md

@@ -1,7 +1,7 @@
 # Pretransforms
 Many models require some fixed transform to be applied to the input audio before the audio is passed in to the trainable layers of the model, as well as a corresponding inverse transform to be applied to the outputs of the model. We refer to these as "pretransforms".
 
-At the moment, `harmonai-tools` supports two pretransforms, frozen autoencoders for latent diffusion models and wavelet decompositions.
+At the moment, `stable-audio-tools` supports two pretransforms, frozen autoencoders for latent diffusion models and wavelet decompositions.
 
 Pretransforms have a similar interface to autoencoders with "encode" and "decode" functions defined for each pretransform.
 
@@ -28,7 +28,7 @@ Example:
 The original [Latent Diffusion paper](https://arxiv.org/abs/2112.10752) found that rescaling the latent series to unit variance before performing diffusion improved quality. To this end, we expose a `scale` property on autoencoder pretransforms that will take care of this rescaling. The scale should be set to the original standard deviation of the latents, which can be determined experimentally, or by looking at the `latent_std` value during training. The pretransform code will divide by this scale factor in the `encode` function and multiply by this scale in the `decode` function.
 
 ## Wavelet pretransform
-`harmonai-tools` also exposes wavelet decomposition as a pretransform. Wavelet decomposition is a quick way to trade off sequence length for channels in autoencoders, while maintaining a multi-band implicit bias.
+`stable-audio-tools` also exposes wavelet decomposition as a pretransform. Wavelet decomposition is a quick way to trade off sequence length for channels in autoencoders, while maintaining a multi-band implicit bias.
 
 Wavelet pretransforms take the following properties: