Expressive Irish Fiddle Performance Model Informed with Bowing

We propose an expressive performance model for celtic fiddle based on the analysis of audio and bowing gestures of real performances. Existing expressivity models deal with perceptual features such as timing deviations or dynamics, but in some cases perceptual features are not enough. We propose a model not only informed with perceptual features but also with bowing gestures which are acquired by means of a 3D motion tracker system. Based on a set of expressive audio features and control gestures we apply machine learning techniques in order to induce an expressive performance model. We use the model to synthesize expressive celtic violin performances from inexpressive score descriptions. In this paper we describe the process of acquiring expressive performance features (both audio and gestures), we detail the automatic performance-score alignment and segmentation, we show how the model is induced, and finally we evaluate the results using a sample based concatenative synthesizer.