Design Ideas for an AI-Driven Pipa Playing Pitch and Rhythm Judgment System

Abstract

Addresses the problem in traditional pipa teaching where pitch and rhythm feedback relies heavily on teachers' on-site listening and is difficult to continuously record during practice. Based on the principle of "assisting teaching rather than replacing evaluation", this paper proposes an AI-driven system for determining pipa pitch and rhythm. The system uses pipa solo audio, standard scores, teacher review comments, and practice feedback requests as primary data, forming a closed loop of "data acquisition—preprocessing—intelligent judgment—score alignment—evidence feedback—continuous correction". The design focuses on summarizing insights from recent core journal research on traditional music datasets, automatic music annotation, time-frequency attention, performance evaluation, and human-computer collaboration. It also considers the characteristics of pipa techniques such as tremolo, strumming, pushing and pulling, vibrato, harmonics, and ornamentation, proposing the system structure, key modules, data processing, deployment route, and feedback boundaries. The system output is not centered on experimental scores but rather on measure location, error evidence, teacher review, and practice suggestions, providing a design reference for the construction of intelligent teaching systems for traditional Chinese plucked string instruments. In terms of pitch accuracy, the accuracy of instrument manufacturing and playing actions are further incorporated into the system logic: the former corresponds to the process calibration of frets, phase and key positions, while the latter corresponds to the left hand's pressing position, string contact angle and string pressing stability, and can be used with the help of machine vision recognition to form action evidence.