karaoke
🎵
Run the lyrics generator
Currently the fine-tuned model lives in BMO. (TODO: make it a service???) Run the following command in the \lyrics directory.
python inference.py new_titles.txt
python postprocess.py
The result can then betrimmed by the ~ artists ~.
Image generation
we use attnGAN and did something similar to what goodkids did here.
The downloaded model also lives in BMO. Commands:
(in karaoke root directory)
python lyrics/generate_images.py
or for a specific song:
python lyrics/generate_images.py --song "hack the fuck out of it"
Midi generation
due to the limited time & compute resource, we are using musegan to generate little pop song bits. (maybe imigrate to magenta's transformer in the future.)
Conda enviornment
For versioning consistency, this repo is designed to be run in a conda env. After cloning, create the env by running:
conda env create -f env-ubuntu.yml
With the right yml file for the operating system. and then
conda activate karaoke
If you installed something, remember to update the yml file by running:
conda env export > env-os.yml
replace "os" with the os you are working in.