AI-enhanced assessment of hybrid musical competencies · Kenyan folk-fusion curricula
The machine measures the signal. The instructor judges the music.
Sauti bifurcates performance assessment into signal-invariant technical evidence, computed transparently, and culturally situated judgement, reserved entirely for human authority. Every divergence between the two is preserved as research data.
Domain 1 · Computed evidence
Signal-invariant measurement
The analysis engine reports only what a signal can attest to without a culturally learned category — with per-metric confidence and self-declared caveats. It is architecturally incapable of producing a grade: its output type has no field for one.
- Pitch stability (monophonic F0 tracking)
- Onset detection & beat alignment
- Tempo induction & local tempo drift
- Timing precision against the beat grid
- RMS energy & dynamic range
- Spectral centroid & waveform views
Domain 2 · Human authority
Cultural judgement, behaviourally anchored
Instructors score the dimensions no signal statistic can carry, against rubrics validated with cultural representatives. The final grade derives from instructor judgement alone, and a representative's determination governs restricted material.
- Improvisational authenticity
- Cultural appropriateness
- Expressive depth
- Musical creativity & interpretation
- Performance communication & ensemble
- Stage confidence
When they disagree, the human decides — and the disagreement becomes data.
Every instructor divergence records the computed value, the human value and a written rationale in an append-only log. That log is not a system feature; it is a primary source of the study's evidence on where automated measurement fails hybrid-genre performance.
What the instrument does
- Assessment workspace
- Waveform, pitch, tempo and energy beside the full rubric — one screen, two clearly separated epistemologies.
- Dual-condition study design
- Every assessment runs in the AI-assisted or the conventional arm, timed for the within-subjects efficiency comparison.
- Cultural governance
- A score at the trigger level on cultural appropriateness withholds the grade and routes the assessment to representative review.
- Research export
- CSV, Excel, JSON and SPSS-ready syntax carrying the raw paired observations, not only the statistics computed from them.
- In-browser signal engine
- Onset, tempo, pitch and energy analysis runs over the Web Audio API on the assessor's own device — inspectable and reproducible.
- Reports with provenance
- Printable reports that keep computed evidence and human judgement visually and semantically distinct.