Omniscient Mozart, being able to transcribe everything in the music.
Project description
omnizart
Omniscient Mozart, being able to transcribe everything in the music, including vocal, drum, chord, beat, instruments, and more. Combines all the hard works developed by everyone in MCTLab into a single command line tool, and plan to distribute as a python package in the future.
A quick-start example is as following:
# Install omnizart
pip install omnizart
# Download the checkpoints after installation
omnizart download-checkpoints
# Now it's ready for the transcription~
omnizart drum transcribe <path/to/audio.wav>
omnizart chord transcribe <path/to/audio.wav>
omnizart music transcribe <path/to/audio.wav>
Or use the docker image:
docker pull mctlab/omnizart:latest
docker run -it mctlab/omnizart:latest bash
Comprehensive usage and API references can be found in the official documentation site.
About
Music and Culture Technology Lab (MCTLab) aims to develop technology for music and relevant applications by leveraging cutting-edge AI techiniques.
Plan to support
Commands | transcribe | train | evaluate | Description |
---|---|---|---|---|
music | :heavy_check_mark: | :heavy_check_mark: | Transcribes notes of instruments. | |
drum | :heavy_check_mark: | :interrobang: | Transcribes drum tracks. | |
vocal | Transcribes pitch of vocal. | |||
chord | :heavy_check_mark: | :heavy_check_mark: | Transcribes chord progression. | |
beat | Transcribes beat position. |
NOTES Though the implementation of training the drum model is 90% complete, but there still exists some invisible bugs that cause the training fails to converge compared to the author's original implementation.
Example usage
omnizart music transcribe path/to/audio omnizart chord transcribe path/to/audio omnizart drum transcribe path/to/audio
For training a new model, download the dataset first and follow steps described below.
# The following command will default saving the extracted feature under the same folder, # called train_feature and test_feature omnizart music generate-featuer -d path/to/dataset # Train a new model omnizart music train-model -d path/to/dataset/train_feature --model-name My-Model
Development
Describes the neccessary background of how to develop this project.
Download and install
git clone https://github.com/Music-and-Culture-Technology-Lab/omnizart.git
# Install dependenies. For more different installation approaches, please refer to the official documentation page.
# The following command will download the checkpoints automatically.
cd omnizart
make install
# For developers, you have to install Dev dependencies as well, since they will not be installed by default.
poetry install
Package management
Uses poetry for package management, instead of writing requirements.txt
and setup.py
manually.
We still provide the above two files for convenience. You can also generate them by executing make export
.
Documentation
Automatically generate documents from inline docstrings of module, class, and function. Hosted document page
Documentation style: Follows numpy
document flavor. Learn more from numpydoc.
Document builder: sphinx
To generate documents, cd docs/
and execute make html
. To see the rendered results, run make serve
and view from the browser.
All documents and docstrings use reStructured Text format. More informations about this format can be found from
Sphinx's Document.
Linters
Uses flake8 and pylint for coding style check.
To check with linters, execute make check
.
You don't have to achieve a perfect score on pylint check, just pass 9.5 points still counted as a successful check.
Caution!
There is convenient make command for formating the code, but it should be used very carefully. Not only it could format the code for you, but also could mess up the code, and eventually you should still need to check the format manually after refacorting the code with tools.
To format the code with black and yapf, enter make format
.
Unittest
Uses pytest
for unittesting. Under construction...
CI/CD
Uses github actions for automatic linting, unittesting, document building, and package release. Currently supports two workflows:
- General check
- Documentation page publishing
- Publish PyPI package and docker image
General Check
Everytime you push to the master branch, file a pull request, and merge into master branch, will trigger this action. This will do checks like code format, and unittests by leveraging the above mentioned tools. If the check fails, you will not be able to merge the feature branch into master branch.
Documentation Page Publishing
We use github page to host our documentation, and is separated as an independent repository.
Please do not directly modify the content of the omnizart-doc repository!!
The only permitted way to update the documentation page is by updating the build_doc
branch, and
let the workflow do the rest of things.
Steps to update the documentation page:
- Clone this repo
- Create a new branch. DO NOT UPDATE THE
build_doc
BRANCH DIRECTLY!! - File a pull request
- Merge into master (by admin)
- Merge into
build_doc
branch (by admin) - Push to this repo (by admin)
Publish PyPI Package and Docker Image
Publish the python package to PyPI and also the docker image to dockerhub when push tags to the repository. The publish process will be automatically done by the github actions. There are several steps in the process:
- Pack and publish the python package.
- Build the docker image and publish to Docker Hub.
- Create release -> this will also trigger the automation of documentation publishment.
Docker
We provide both the Dockerfile for local image build and also the pre-build image on Docker Hub.
To build the image, run the following:
docker build -t omnizart:my-image .
To use the pre-build image, follow below steps:
# Download from Docker Hub
docker pull mctlab/omnizart
# Execute the image
docker run -it mctlab/omnizart:latest
### For those who want to leverage the power of GPU for acceleration, make sure
### you have installed docker>=19.03 and the 'nvidia-container-toolkit' package.
# Execute the docker with GPU support
docker run --gpus all -it mctlab/omnizart:latest
Command Test
To actually install and test the omnizart
command, execute make install
. This will automatically create a virtual environment and install everything needed inside it. After installation, just follow the instruction showing on the screen to activate the environment, then type omnizart --help
to check if it works. After testing the command, type deactivate
to leave the virtual environment.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for omnizart-0.1.0b2-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 23ad5796d0784f72db21ce6c6ef4bbf05d2cf885aa20331100557c22e68379c1 |
|
MD5 | 8a994bed0d82e82639d019a6fdf939c8 |
|
BLAKE2b-256 | 805db0563159f59b141196535cd01bc841240bc9802dab10d8a8ea78521870b9 |