CUEX: An algorithm for automatic extraction of expressive tone parameters in music performance from acoustic signals

摘要

CUEX is an algorithm that from recordings of solo music performances extracts the tone parameters for tempo, sound level, articulation, onset velocity, spectrum, vibrato rate, and vibrato extent. The aim is to capture the expressive variations in a music performance, rather than to identify the musical notes played. The CUEX algorithm uses a combination of traditional methods to segment the audio stream into tones based on fundamental frequency contour and sound level envelope. From the resulting onset and offset positions, the different tone parameters are computed. CUEX has been evaluated using both synthesized performances and recordings of human performances. For the synthesized performances, tone recognition of 98.7% was obtained in average. The onset and offset precision was 8 ms and 20 ms, respectively, and the sound level precision about 1 dB. Various applications of the CUEX algorithm are discussed. For human performances, the recognition was 91.8% in average.