Part 2: The Data — Building the First Public Coffee Roasting Audio Dataset with Warp/Oz

There is no widely available public audio dataset for coffee roasting first crack detection. I was not able to find one on Hugging Face, Kaggle, or in the literature. Before the model in Post 1 could train a single epoch, I had to build one from scratch: recording sessions, annotating audio in Label Studio, and designing a pipeline that avoids a common failure mode in time-series audio ML where evaluation metrics are quietly inflated. The result is a publicly available coffee roasting audio data