Notebooks

The notebooks/ folder holds a numbered tutorial set. Each notebook runs on synthesized audio when no dataset is configured, so a fresh clone needs only pip install "afx-audio[notebooks]", plus "afx-audio[torch]" for 06.

Notebook	Covers
01 Quickstart	load_audio, configs, extract_all_features, flatten_features, preset ids (speech@1), the feature catalogue with its undefined and torch entries, config validation
02 Frames and aggregation	per-frame arrays and shared frame times, plots, NaN frames, coverage and <feature>_valid_frac, aggregation methods
03 Long recordings and custom features	extract_segments, register_feature, worker processes
04 Batch and parallel	extract_features_from_folder, output formats, failed files and AFXWarning, provenance, dataset statistics and scaler_from_stats, n_jobs, the afx command
05 scikit-learn and selection	AFXFeatureExtractor, config='speech@1' and provenance_, cross-validation, grid search over extraction settings, feature selectors
06 PyTorch	afx_audio.torch: parity, batches, gradients, AFXFeatureModule
07 Domain presets	a preset id end to end on its domain dataset, with the config hash behind the result
08 eGeMAPS	eGeMAPS v02 on synthetic voices with known F0, jitter and shimmer, the 88 functionals and 25 LLDs, the egemaps@1 preset in a pipeline, an optional openSMILE comparison
Benchmark validation	agreement with librosa, torchaudio, openSMILE, spafe and Praat

Dataset environment variables and the test runner are described in notebooks/README.md.