Workflow
- Combine all directory recordings to single file:
recordings/mp3] $ sox 2020-04-04/*mp3 /output/path/daily-2020-04-04.mp3
- Move single file to js manual classifier:
mv daily-2020-04-04.mp3 /path/to/js-audio-classifier
- Classify audio:
cd /path/to/js-audio-classifier vim app.js # set filename ./start-webserver.sh
-
Save exported JSON to /output/path/daily-2020-04-04.json
-
Save mp3 there and convert to WAV for speed:
ffmpeg -i daily-2020-04-04.mp3 daily-2020-04-04.wav
- Combine all exports to a single set:
Add new set to CLASSIFICATION_SETS in extract_data.json
cd /output/path; python3.6 extract_data.py
- Train NN on that single set:
python3.6 build_nn.py
Prediction
python3.6 predict.py [model filename] [audio filename] [audio start position] python3.6 predict.py model-2020-04-08_16:31:10.h5 daily-2020-04-04.wav 321