Research Goal: To develop intelligent systems capable of automatic recognition of audio content. Piano Transcription: Music transcription is the process of creating a musical score from an audio recording. We've created an automatic piano transcription system that learns to identify the notes present in complex, polyphonic recordings based only on labeled training examples. EURASIP Journal on Advances in Signal Processing article (9pp) Piano transcription training examples Posteriorgram (note prediction probability) Melody Transcription: The melody of a piece of music provides a convenient and natural description for summarization, communication, and retrieval. Just like our approach to piano transcription, we've developed an automatic melody transcription system that learns to identify the melody in complex, polyphonic music (e.g. pop songs). Machine Learning Journal article (18pp) ISMIR05 Proceeding (6pp), Slides (18pp) MIREX05 Abstract (3pp) Melody transcription training examples Posteriorgram (note prediction probability) SVM Active Learning for Music Retrieval: How do you define music similarity? Mike Mandel and I built a music retrieval system that conforms to each user's definition of audio similarity. The user shows the system an example of what he/she's looking for, then the system identifies acoustically similar songs through a few short rounds of feedback. Imagine using something like this to make a playlist or to find new music. Playlist Generation Demo Multimedia Systems Journal article (10pp) Automatic Playlist Generator

HOME   PUBLICATIONS