Aadi: automatic Konnakol transcription

Konnakol is the South Indian art of vocal percussion: performing rhythm as spoken syllables. Aadi is a research project asking whether a machine learning system can transcribe Konnakol audio into structured, timestamped rhythmic syllables.

The project is built as a reproducible research package, working incrementally: define the rhythmic vocabulary and problem formulation, build an annotated dataset, establish a baseline ASR model, then add onset-based segmentation, syllable classification, and rhythm-grammar correction on top. The target outputs are a working transcription system and a publishable research package with dataset documentation, experiments, and evaluation scripts.

It sits at the intersection of the two halves of this site: machine learning engineering applied to a rhythm tradition I care about as a musician. Still in progress; the honest status is “dataset and problem formulation,” not “results.”