Expressive Performance Modeling in Polyphonic Flamenco and Jazz Guitar

In musical performances, expressive models have been proposed in order to study the analysis and characterization of the deviations that a musician introduces when performing a musical piece. In this work, expressiveness in guitar performance is presented as a case of study. However, in guitar case studies, the ability to produce polyphonic sounds and its musical versatility, results on a variability of sounds, possessing significant challenges on the automatic transcription and on the analysis of the recorded data when building the dataset for the model. Data-driven methods have become increasingly popular in order to approach problems like guitar transcription and, datasets of labeled audio data are highly needed. This work proposes a system capable of collecting audio and MIDI annotations from guitar performances based on specific musical pieces, jazz and flamenco in this work. In particular, by recording guitar performances using a hexaphonic pickup, the system is able to, not only provide recording of the individual strings, but also to largely automate the annotation process of the expressive content of the music performance. Then a pretrained machine learning model is used in order to automatically generate music sequence excerpts based on the acquired data. Finally, di˙erent sets of data are presented in order to compare results and qualitative evaluations of the generated sequences are performed