Using Concatenative Synthesis for Expressive Performance in Jazz Saxophone

We present here a concatenative sample-based saxophone synthesizer using an induced performance model intended for expressive synthesis. The system consists on three main parts. The first part provides the analysis of saxophone expressive performance recordings and the extraction of descriptors related to different temporal levels. With the obtained descriptors and the analyzed samples, we construct an annotated sample database extracted directly from the performances. For the second part, we use the annotations to induce a performance model capable of predicting some features related to expressivity. In the third part, the predictions of the performance model are used to retrieve the most suitable note samples for each situation, and transform and concatenate them following the input score and the induced model.