Against the backdrop of national initiatives in aesthetic education and education digitalization (informatization) in China, music education requires robust assessment instruments. Anchored in the Technological Pedagogical Content Knowledge (TPACK) framework and attuned to the performative, compositional, and practice-oriented nature of music education, this study offers a localized, discipline-specific operationalization of TPACK. Focusing on pre-service music teachers at universities in Beijing, we assess TPACK levels, elucidate mechanisms of technology-content-pedagogy integration, and evaluate instructional effects. A mixed-methods design was employed, comprising scale revision and survey administration (N = 554), structural equation modeling (SEM) and random forest analyses, supplemented by classroom observations, interviews, and a field-based teaching experiment. The instrument demonstrated strong psychometric properties (KMO = 0.945; Cronbach’s α across dimensions > .970; standardized α = .976), and SEM indicated acceptable model fit (CFI = 0.917; RMSEA = 0.077). Significant paths included PK → TPK, PK → PCK, CK → TCK, CK → PCK, and TK → TPK (with CK → TCK the strongest). TPCK was significantly predicted by TPK (strongest), TCK, and PCK, whereas the direct paths from TK, CK, and PK to TPCK, as well as TK → TCK, were non-significant. The random forest model showed that out-of-bag error decreased and then plateaued as the number of trees increased, with overall accuracy of 83.58%, a tendency to classify cases into “category 5,” and variable importance highest for PCK and CK. In the field-based experiment, the experimental group outperformed the control group on the four dimensions of Keller’s ARCS (Attention, Relevance, Confidence, Satisfaction) by 0.164-0.194 units and achieved a 2.6% gain in achievement scores. Taken together, these findings suggest that a localized measurement of TPACK can provide a reliable basis for diagnosing pre-service teachers’ competencies and informing curriculum reform, and that deep integration of technology with content knowledge is pivotal for instructional improvement. Limitations include sampling confined to Beijing and limited granularity within the technological knowledge dimension; future work should broaden sampling, incorporate AI-related competencies, and conduct cross-cultural comparisons.