Machine Learning of Personal Gesture Variation in Music Conducting
summary
A system learns idiosyncratic gesture variations to drive articulation in music conducting using a Gaussian Mixture Model that maps user-provided examples (three points from legato to staccato) into a continuous 1–3 control value in real time. A user study with 10 musicians and 10 non-musicians yields mean accuracy 0.31 and a correlation 0.6 between accuracy and perceived control, shows musicians outperform non-musicians, finds training separability does not predict performance, and reveals high cross-subject error (idiosyncrasy) and benefit from real-time feedback.
AI-written summary · about
chapters (5)
- 0:00Research motivation
- 1:33Experimental methodology
- 3:29Training and performance
- 5:57Results and analysis
- 8:36Conclusion and future work
related talks
41mCreating Intelligent Music Instruments with Machine Learning
34mModelling emotional nuance of musical performance
05mPositioning left-hand movement in violin performance: system and user study on fingering pattern ...
17mMuEns: A Multimodal Human-Machine Music Ensemble for Live Concert Performance
22mSeeing, Sensing and Recognizing Laban Movement Qualities
33mContext-Aware Masking for Controlled Music Generation in MIDI
34mThe hand prosthesis that learns from the user
05mInteractive Exploration-Exploitation Balancing for Generative Melody Composition