To run the web application locally you need to do two main steps:
- Install Node.js and npm packages to run the web application
- Install python and python packages to run audio generation
First, you need to download and install Node.js on your computer. Next clone the repository:
git clone https://github.com/vagechirkov/wt21-the-noise-enricher.gitLastly, run the following commands in the terminal:
cd node_app/server/
npm install
nodemon app.jsThen you should be able to open the application in your web browser: http://127.0.0.1:3030/
⚠️ Note: runnpm installcommand from the node_app/server/ directory.
The next step is to install python and python packages. This is necessary to run the audio generation process.
One way to manage python packages is to use Anaconda. You can download it
here: Anaconda. After installing it, you can install the required python packages
from the requirements.txt file using pip. Here are the commands to type in the terminal:
conda create -n noise-enricher python=3.8
conda activate noise-enricher
pip install -r requirements.txtFor this project, it was decided for the sake of simplicity and convenience to keep AI-related processing in the same app container (“Dyno”) on the Heroku server. The main benefit of this approach is that the audio files uploaded by the user and generated by the AI algorithms are immediately accessible for both AI and frontend since these files are on the same server. Heroku allows us to do that using several buildpacks in the same application. Here is the list of all buildpacks that we used:
First major part of our setup is the heroku/nodejs buildpack that is responsible for the Node.js server. It uses
the package.json file in the root of the GitHub repository. This file includes commands to build and
start the express server from the node_app/server folder:
"scripts": {
"heroku-prebuild": "cd node_app/server && npm install && npm run build",
"start": "cd node_app/server && npm start"
}The second main component of our setup is the heroku/python buildpack that is responsible for the python environment.
It requires requirements.txt and runtime.txt files in the root of the GitHub
repository. Although Heroku allows to install any pipinstallable python package, there is an important limitation that
is the “Slug size” which amounts to 500MB. See more
information here
. One can significantly reduce the slug size for Pytorch library
using this method
and this method
for TensorFlow. There are also
general recommendations on Heroku
how to decrease the slug size. But both main machine learning libraries ( Pytorch and TensorFlow) cannot be deployed
in the same web application on Heroku. The final set of the installed python packages is the following:
tensorflow-cpu==2.7.0
librosa==0.8.1
AudioSegment==0.23.0
pydub==0.25.1
h5py==2.10.0
pysoundfile==0.9.0
Additionally, to run python package SoundFile correctly we
installed heroku/heroku-buildpack-apt. One can find more
information about the error and the
solution here. This buildpack needs
the Aptfile in the root of the GitHub repository which has the following content:
libsndfile1
libsndfile1-dev
After setting up all the configuration files, one can simply follow the description of
the git-based deployment process. After this on can simply push develop
branch to Heroku main brand:
git push heroku develop:mainHere is how the successful build and deployment looks like from the Heroku CLI:
UX
WD
AI