Automatic Drum Transcription (ADT) describes the process of transcribing audio files containing drum sounds into a musical representation indicating the onsets of each drum sound in the recording.
This project is carried out in collaboration with Drumstik, and aims at producing automatically a musical score from a real-time audio recording.
For this application, I use a deep learning approach. This model compute a probability at all times that is use to know if some parts of the drum (snare, bass, hi-hat, etc.) are played.