Audio Classification • 86.2M • Updated • 6
Coffee First Crack Audio Dataset
Audio dataset for training coffee first crack detection models. Contains 10-second WAV chunks from coffee roasting recordings, labelled as first_crack (popping/cracking sounds) or no_first_crack (background roast noise).
Model: syamaner/coffee-first-crack-detection Source code: github.com/syamaner/coffee-first-crack-detection
How this was built:
- Part 1 — The Architecture & The Agent — Spec-Driven ML Development with Warp/Oz
- Part 2 — Building an Open Coffee Roasting Audio Dataset with Warp/Oz
- Part 3 — The Science — Hyperparameter Tuning to 100% Precision with Warp/Oz
- Part 4 — The Edge — Deployment of an 86M Parameter Audio Transformer
- Part 5 — From Local Model to Live Demo — Publishing to Hugging Face with Warp/Oz
Original prototype:
Dataset Summary
- 973 total chunks (fixed 10-second sliding windows, no overlap)
- 15 source recordings from 2 microphones, 3 coffee origins
- Recording-level split (no data leakage between splits)
- 20% first_crack / 80% no_first_crack — realistic class imbalance
| Split | first_crack | no_first_crack | Total | Recordings |
|---|---|---|---|---|
| Train | 124 | 463 | 587 | 9 |
| Val | 37 | 158 | 195 | 3 |
| Test | 36 | 155 | 191 | 3 |
Annotation Approach
Each source recording was annotated in Label Studio with a single first_crack region spanning from the first audible pop to the end of consistent cracking. The chunk_audio.py script then slid fixed 10-second windows across each recording and labelled each window based on overlap (>=50% threshold) with annotated first_crack regions.
This approach replaces the prototype method of manually annotating 20-30 small regions per file, producing consistent real-audio training chunks that match what the model sees during inference.
Features
| Feature | Type | Description |
|---|---|---|
audio |
Audio (16kHz) | 10-second mono WAV chunk |
label |
string | first_crack or no_first_crack |
label_id |
int | 1 = first_crack, 0 = no_first_crack |
microphone |
string | mic-1-original or mic-2-new |
coffee_origin |
string | e.g. brazil, costarica-hermosa, brazil-santos |
Source Recordings
| Mic | Origin | Recordings | Notes |
|---|---|---|---|
| mic-1-original | costarica-hermosa | 5 | Legacy recordings from prototype |
| mic-1-original | brazil | 4 | Legacy recordings from prototype |
| mic-2-new | brazil | 4 | New recordings (Feb 2026) |
| mic-2-new | brazil-santos | 2 | New recordings (Apr 2026) |
Usage
from datasets import load_dataset
ds = load_dataset("syamaner/coffee-first-crack-audio")
print(ds)
# DatasetDict({
# train: Dataset({features: [audio, label, ...], num_rows: 587})
# val: Dataset({features: [audio, label, ...], num_rows: 195})
# test: Dataset({features: [audio, label, ...], num_rows: 191})
# })
# Access a sample
sample = ds["train"][0]
print(sample["label"], sample["microphone"], sample["coffee_origin"])
Citation
@misc{yamaner2026coffeefc,
author = {Yamaner, Sertan},
title = {Coffee First Crack Audio Dataset},
year = {2026},
url = {https://huggingface.co/datasets/syamaner/coffee-first-crack-audio}
}
- Downloads last month
- 69