What it does
It takes in Jazz sound files inputted through a web interface and produces the sound file with a Hip Hop cadence. This is also easily applied to other genres.
How we built it
We were inspired by this paper https://junyanz.github.io/CycleGAN/ which translates images from one source domain X to those of another target domain Y, and decided to apply the same technique to music genres.
Users can upload sound files in our website. Files are uploaded to our flask server where an HTTP request for a transformed file is sent to our ML Model stored in the Google Machine Learning Engine. The transformed sound file prediction is then displayed on our site.
Challenges I ran into
Understanding and implementing the Google Cloud Platform workflow was a major hurdle at the beginning but we were able to connect the ML Engine with our backend. Building the tensorflow model and using the Google ML Engine to train it were also big challenges.