song-recognition

song-recognition copied to clipboard

Pre Requirements

• ffmpeg (required by librosa to load mp3)
• nodejs (frontend)
• MongoDB

Environment Setup

Inside backend/.env:

Edit the default MONGO_URI to your mongodb connection key

MONGO_URI=mongodb://localhost:27017/?readPreference=primary&serverSelectionTimeoutMS=2000&appname=MongoDB%20Compass&directConnection=true&ssl=false
FLASK_ENV=development

Backend

1. Create conda env (optional)

conda create --name kishikan python=3.8

conda activate kishikan

2. Install python dependencies (in backend)

Please navigate to backend folder, then

pip install -r requirements.txt

3. Configure flask (in backend)

export FLASK_APP=app

start the flask using flask run

Frontend

Please navigate to frontend folder, then

yarn install to install dependencies

yarn start to open client

Important Folders Structure

.github
-- postman collection for API testing
backend
-- app/: all code to implement RESTful Flask Server
-- kishikan/: audio fingerprinting module
-- nazo/: query by humming module
frontend: react frontend code in typescript

Demo

Please open backend/audio_fingerprinting.ipynb for audio fingerprinting, and backend/query_by_singing.ipynb for query by humming.

Experiments

If you want to perform experiments, please download the datasets and place them into datasets/:

GTZAN and it's query: https://www.music-ir.org/mirex/wiki/2021:Audio_Fingerprinting

QBSH midi and query: https://www.music-ir.org/mirex/wiki/2021:Query_by_Singing/Humming

Benchmark usage can be found inside jupyter notebooks in backend/