Music Retrieval Systems

Course 3 builds upon the digital signal processing concepts we have learned in Course 1 and the machine learning concepts we have learned in Course 2 to investigate a variety of interesting music information retrieval tasks. As these tasks become more advanced and complicated, the examples and assignments in this course shift from programming examples from scratch to utilizing existing libraries and frameworks. Topics explored include: music recommendation and query-by-humming, automatic chord detection and cover song identification, automatic music transcription and sound source separation, and audio fingerprinting and watermarking. By completing the course, you will have a good coverage of all the work that has been done in the field of MIR.

• Query Retrieval
• In this session, we will cover the basics of audio fingerprinting and watermarking: audio landmark extraction, quantization, jaccard similarity, minhash, locality sensitive hashing.
• Audio Fingerprinting and Watermarking
• Principal component analysis, self-organizing maps, visualization in MIR will be covered in this session.
• Transcription and Sound Source Separation
• In this session, we will cover MIDI, symbolic music representations, dynamic programming, self-similarity matrices, polyphonic audio-score alignment.
• Polyphonic Alignment and Structure Segmentation
• This session will describe the problem of chord detection, a quick introduction to music theory and notation, hidden markov models and other types of probabilistic modeling for chord detection and structure segmentation.
• Chord Detection and Cover Song Identification
• We will learn about the basic architecture of a query-by-humming system, theme extraction, note segmentation and quantization.