Adding some more docs and READMEs

PiperOrigin-RevId: 292658348
Change-Id: I281f21394a3391f39fb8cf29eb72c5248e769585

README.md

@@ -276,7 +276,7 @@ If you use this code please cite it as:
 ```latex
 @inproceedings{
   engel2020ddsp,
-  title={{\{}DDSP{\}}: Differentiable Digital Signal Processing},
+  title={DDSP: Differentiable Digital Signal Processing},
   author={Jesse Engel and Lamtharn (Hanoi) Hantrakul and Chenjie Gu and Adam Roberts},
   booktitle={International Conference on Learning Representations},
   year={2020},

ddsp/colab/README.md

@@ -0,0 +1,8 @@
+# Colab Notebooks
+
+Interactive notebooks to demonstrate DDSP.
+
+* [demos](./demos/): Self-contained demonstrations for training models and showing them in action.
+
+* [tutorials](./tutorials/): Interactive walkthroughs of the DDSP functions and APIs.
+

ddsp/training/README.md

@@ -8,6 +8,11 @@ The current supported models are variants of an audio autoencoder.
 <img src="https://storage.googleapis.com/ddsp/additive_diagram/ddsp_autoencoder.png" width="800px" alt="logo"></img>
 </div>
 
+# Disclaimer
+*Unlike the base `ddsp/` library, this folder is actively modified for new
+experiments and has a higher chance of making breaking changes in the future.*
+
+
 ## Modules
 
 The DDSP training libraries are separated into several modules:
@@ -126,9 +131,3 @@ ddsp_run \
 --gin_param="batch_size=16" \
 --alsologtostderr
 ```
-
-
-# Disclaimer
-*Unlike the base `ddsp/` library, this folder is actively modified for new
-experiments and has a higher chance of making breaking changes in the future.*
-

ddsp/training/data_preparation/README.md

@@ -0,0 +1,3 @@
+# Data Preparation
+
+Scripts and libraries to prepare datasets for training.

ddsp/training/gin/datasets/README.md

@@ -0,0 +1,5 @@
+# Gin Configs: Datasets
+
+This directory contains gin configs for datasets. Each file contains datasets to use for training, evaluation, and sampling.
+Each training run with `ddsp_run.py` must be supplied both a model file and a dataset file, each with the `--gin_file` flag.
+Evaluation and sampling jobs only require a dataset file.

ddsp/training/gin/models/README.md

@@ -0,0 +1,5 @@
+# Gin Configs: Models
+
+This directory contains gin configs for model architectures (including ddsp modules).
+Each training run with `ddsp_run.py` must be supplied both a model file and a dataset file, each with the `--gin_file` flag.
+Evaluation and sampling jobs only require a dataset file.

ddsp/training/gin/optimization/README.md

@@ -0,0 +1,4 @@
+# Gin Configs: Optimization
+
+This directory contains gin configs for training hyperparameters such as `batch_size`
+and `learning_rate`. Default values are loaded via `ddsp_run.py` and can be replaced with the `--gin_param` flag.

ddsp/training/gin/papers/README.md

@@ -0,0 +1,21 @@
+# Gin Configs: Papers
+
+This directory contains gin configs for replicating the experiments in published
+papers using ddsp. Each config file specifies both the dataset and the model,
+so only a single --gin_file flag is required.
+
+
+## List of papers
+
+* [iclr2020](./iclr2020/): Experiments from the orginal DDSP ICLR 2020 paper ([paper](https://openreview.net/forum?id=B1x1ma4tDr), [blog](https://magenta.tensorflow.org/ddsp)).
+
+```latex
+@inproceedings{
+  engel2020ddsp,
+  title={DDSP: Differentiable Digital Signal Processing},
+  author={Jesse Engel and Lamtharn (Hanoi) Hantrakul and Chenjie Gu and Adam Roberts},
+  booktitle={International Conference on Learning Representations},
+  year={2020},
+  url={https://openreview.net/forum?id=B1x1ma4tDr}
+}
+```