Computational Detection of Local Cadence on Revised TPS

Cadential Retention is a process to combine the V-I progression to regard as a single pitch event and to give a high salience, in the time-span tree of the Generative Theory of Tonal Music. Though the theory was implemented as an automatic analyzer, this process has been lacked and needed to be compensated. In this paper, we propose a fundamental procedure to realize this. We presuppose that the chord names are already assigned on the target score; then, the system first hypothesizes all the possible combinations of the key and the degree, finds plausible connections between them, and detects cadences. We employ dynamic programming to connect chords, and calculate the chord distance by the revised Tonal Pitch Space. Also, we restrict the minor scale to the harmonic one to avoid the ambiguity in chord interpretation. In our system, we show the final result of the revised time-span tree, including the local cadences.