-
-
Poster
Inspiration
We came to this topic after seeing some projects that attempted to classify musical genres and then learning about methods for language translation in class. If deep learning models can translate text from one language to another, it seems like a logical next step to try translating audio from one "language" to another. We felt that the analogue in audio would be converting music from one genre to another. There is, however, a key difference between this problem and language translation: we do not have paired data. The dataset does not have examples of the same song in different genres. As such, our model will have to learn by itself the features that identify a genre.
Link to check-in reflection PDF
Link to music survey where you can hear sample outputs
Original (midi files that went through preprocessing and postprocessing, but didn't go through the model):
moonlight sonata autumn leaves
ResNet:
moonlight sonata (jazz) autumn leaves (classical)
LSTM V2:
Built With
- python
- tensorflow
Log in or sign up for Devpost to join the conversation.