/TransformerADT

Deep Learning for Audio and Music assignment: A vanilla transformer encoder that does Audio Drum Transcription (ADT) on the Groove Dataset and a VAE to generate synthetic training data

Primary LanguagePython

This is the repo accompanying my ECS7013P Deep Learning for Audio and Music coursework assignment submission: Automatic Drum Transcription with Data Augmentation.

You can clone this repo and its submodules using the following command

git clone --recurse-submodules -j8 https://github.com/pelinski/TransformerADT.git

In order to run the code, you can use the provided environment:

conda env create -f environment.yml

and activate it using

conda activate groove

For some reason, this usually gives errors, so alternatively you can create your own environment with python 3.6

conda create --name groove python=3.6
source activate groove

and install the following packages:

pip install visual_midi
pip install tables
pip install magenta==1.1.7 --use-deprecated=legacy-resolver
pip install note_seq

conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch -c conda-forge
pip3 install torch==1.10.2+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
conda install -c conda-forge fluidsynth
pip install pyFluidSynth
pip install wandb
conda install -c anaconda wget
pip install bokeh
pip install pandas
pip install PySoundFile
pip install colorcet
pip install holoviews
pip install ipyparams

You will also need the datasets and evaluators, that you can get from here: