The Noise Enricher

It's the opposite of noise cancellation! Read more in our Blog Post.

Repository structure

• UX materials and links
• AI scripts
• WD files

Steps to run the prototype locally

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

Node.js

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.git

Lastly, run the following commands in the terminal:

cd node_app/server/
npm install
nodemon app.js

Then you should be able to open the application in your web browser: http://127.0.0.1:3030/

⚠️ Note: run npm install command 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.

Python

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.txt

HEROKU deployment

For 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:main

Here is how the successful build and deployment looks like from the Heroku CLI:

Dependencies

Node.js

• Express
• Embedded JavaScript templates
• Multer
• nodemon
• connect-flash
• ejs-mate

JavaScript/CSS

• wavesurfer.js
• Bootstrap
• JQuery

Python

• Pytorch
• SoundFile
• TensorFlow
• Librosa

Authors

UX

• Yulia (GitHub, LinkedIn)
• Marc (GitHub)

WD

• Gianna (GitHub, LinkedIn)
• Valerii (GitHub, LinkedIn)

AI

• Christian (GitHub, LinkedIn)
• Raphael (GitHub, LinkedIn)
• Seungcheon