An Interactive Singing Feedback System for Musicians: A Mobile Application for Automated Pitch Detection and Personalised Vocal Training
Authors: ADEGBITE Racheal Modupe, OLALEKAN Isaac Damilola, ADEGBITE Oluwaseyi, RASHEED Sofiat Adeola, GBAYE Eunice Abisola, CHUKWU Emmanuel AruchiThe improvement of musicians depends on voice training and singing instructions but conservative instruction is restricted by insufficient number of qualified coaches and high cost of engaging them, delayed response, broad lesson content that downplay unique vocal profiles. For upcoming singers in remote and underdeveloped areas, these hindrances are aggravated. A mobile-enabled interactive singing feedback system that gives near-instant pitch detection, key detection, customized AI-generated vocal feedback mechanism was designed, implemented and evaluated in this study. WebRTC Voice Activity Detection was used for noise detection, Librosa for audio feature extraction while for the application, React Native and Expo was used for the frontend and Flask backend.mIn order to assess the usability and applicability of feedback, system efficiency testing was combined with user trials. The system responds quickly: pitch and key feedback derived from on-device and backend signal processing is returned in approximately 0.47 seconds, with the additional natural-language coaching from the cloud language model arriving shortly afterwards. The system achieves 92% accuracy in pitch detection against reference tones, while voice activity detection removes 85% of non-vocal segments. The results revealed that integrating mobile technology and machine learning into vocal pedagogy can provide a personalized, economical, and alternative to conventional training, which is useful most especially to students.

