DDSP demo on Streamlit

Streamlit version for the DDSP timbre transfer demo but only for .wav files.

Prerequisites

• You need to download and unzip the DDSP checkpoints in the models/ folder.

I did it using the gsutil tool in a Colab environment then downloading from Colab to local machine :

!gsutil cp -r gs://ddsp/models/ .

The structure should look like :

├── models
│   ├── solo_flute_ckpt
|   |   ├── model.ckpt-2400.data-00000-of-00001
|   |   ├── model.ckpt-2400.index
|   |   ├── model.ckpt-2400.meta
|   │   └── operative_config-0.gin
│   ├── solo_flute2_ckpt
|   │   └── ...
│   └── slow_violin_ckpt
|       └── ...
│
├── app.py
│
.
.
.

• You also need some .wav file to upload. There is no means to record your voice yet within Streamlit so do it beforehand, or download some free voice sample.

Install

I use conda but feel free to use your own virtual env to install requirements.txt :

conda create -n ddsp-streamlit-ui python=3.7
conda activate ddsp-streamlit-ui
pip install -r requirements.txt

Run

streamlit run app.py

At the end of script, the output file is created in output.wav.

Contribute

To regenerate requirements.txt :

pip install pip-tools
pip-compile requirements.in

To format code :

pip install black reorder-python-imports
reorder-python-imports --application-directories=.
black app.py