An interactive Deep Learning based Flask Web App that records audio and predicts who is the singer.
- Install all the requirements
- In command line, python main.py
- Copy the URL and paste it on your browser
https://drive.google.com/open?id=1HTtM8yIG__YciGv-JbVjlh7C3kfKtM5G
Dataset was created by downloading 40 songs as .wav each of 6 different singers, creating 5 clips of each song where probabibilty of singing is relatively high, converting .wav file into spectrograms.
6 Classes named:
- Arijit Singh
- Avril Lavigne
- Ed Sheeran
- Lata Mangeshkar
- Sonu Nigam
- Taylor Swift
After 20 epochs(1hr training) - Testing accuracy~64%
- As audio is a spatio-temporal data, it is vey hard to represent this time-dependent on a fixed scale. So, it is obvious that accuracy is relatively less.
- Flask 1.1.2
- Tensorflow 2.0.0
- Keras 2.3.1
- Scipy 1.4.1
- Pillow 7.1.1