Build notes
What I learn while building, written down.
Engineering notes on music retrieval and audio AI, written next to the code: what a representation keeps and what it discards, how key and tempo invariance are actually implemented, where a baseline breaks, and which evaluation choices decide the result.
I write them so the path behind a working system is visible enough for someone to inspect, and to argue with.
Folders
Recent Posts
-
Audio Explorer: Building the Audio AI Pipeline from First Principles
A visual, first-principles tour from decoded waveform to STFT, Mel spectrograms, MFCCs and beat tracking: the feature pipeline in front of modern audio AI.
-
Why chroma is the right representation for cover detection
Working notes on the classical baseline: what the Optimal Transposition Index assumes, why subsequence DTW absorbs tempo drift, and how easily an unnormalised alignment score flatters short candidates.