Automated detection of legato transitions in clarinet performance using enhanced second derivative analysis

Accurate identification of note-to-note transition types in wind instrument performance is essential for automated music analysis, performance assessment, and music technology applications. Traditional threshold-based methods relying on peak steady-state pressure often fail to capture subtle articulations. While curvature-based approaches using the second derivative of rms mouthpiece pressure show promise, they tend to mischaracterize smooth legato transitions that lack distinct pressure variations. This work presents an enhanced extrema-based model for automatically detecting clarinet transition durations. The method defines ΔT as the time between salient extrema in the second derivative signal. While earlier implementations using a valley-to-valley approach worked well for portato transitions, a refined definition—spanning from the first to last detected extrema—offers improved detection of legato transitions with minimal pressure variation. The fully automated system operates without user input or manual tuning and performs robustly across articulation types. Its flexibility and reliability make it well suited for real-time performance feedback, pedagogical tools, and broader music technology applications.