Audio Unmixing and Music Score Transcription. Developed as my undergraduate final year project.
conda create -n aumix-test-venv python=3.6 anaconda
.Install aumix
as a package by pip install --index-url https://test.pypi.org/simple --no-deps aumix-DoraMemo
.
Alternatively, append the aumix
folder in the cloned repository to the sys.path
variable:
import sys
sys.path.append("path_to_aumix")
pip install -r requirements.txt
.Change the directory to the script you want to run, then run it by python3 <script_name>
. For example:
repo$ cd examples
repo$ python3 fourier_transform.py
This is done to preserve the relative paths used in some scripts, such as examples/eval/separation/urmp_gen.py
.
The URMP dataset has to be downloaded separately due to its size. It is available at http://www2.ece.rochester.edu/projects/air/projects/URMP.html. Although we will only use the audio files, the whole dataset (12.5GB) has to be downloaded by filling out the Google Form at https://goo.gl/forms/xSvMzlwl3IWijvcp2. The video files can be subsequently deleted.
For the evaluation scripts to work automatically, the URMP dataset has to be placed in the data
folder with this structure:
repo/
└── data/
└── URMP/
└── Dataset
└── 01_Jupiter_vn_vc
└── AuMix_01_Jupiter_vn_vc.wav
└── AuSep_1_vn_01_Jupiter.wav
...
└── 02_Sonata_vn_vn
...
The sample piece, 34_Fugue_tpt_tpt_hn_tbn, has been included in the repository as a guide.
The size of the plots in the visualiser might not be the intended size, since an IDE was used during the development and the plots show up fine in the plots pane. In any case, saving the plots should retain the correct size. A figure can be saved by passing the savefig_path
parameter in aumix.plot.plot.single_plot
and aumix.plot.plot.single_subplots
. For example:
import aumix.plot.plot as aplot
# Create FigData
aplot.single_plot(fig_data=fig_data,
savefig_path="output.png" # <<< this
)