Inspired by a love for music transcription, we decided to explore the realm of music information and signal processing to feel out the limits of the field. Our goal is to assist in the transcription of all forms of music for performance enthusiasts and for historical preservation.

What it does

Accepts digital audio files and attempts to decompose individual pitches for human analysis.

How we built it

Examples are arranged into Jupyter Notebooks for presentation. Processing utilizes Librosa for signal processing and TensorFlow for machine learning approaches.

Challenges we ran into

Our team was relatively inexperienced in Music Information Retrieval and Machine Learning concepts going into this project, and there was a lot to learn before meaningful work could be done.

Accomplishments that we're proud of

Successful decomposition of simple monophonic melodies and relative success with polyphony.

What we learned

We learned many core concepts in MIR and ML.

What's next for Audio Signal Processing

We hope to implement more machine learning features in the future, including neural nets etc..

Built With

  • librosa
  • magenta
  • python
  • tensorflow
Share this project: