""" Builds the training-distribution reference file that scripts/drift_check.py compares live traffic against. Run this once per training set / after each retrain — not continuously. Usage: python scripts/build_drift_reference.py --data-csv data/labels.csv \ --audio-col path --out data/drift_reference/train_features.json """ import argparse import json from app.audio.decode import decode_audio_to_tensor, to_mono_16k from app.audio.validate import validate_and_extract_features import pandas as pd def main(): ap = argparse.ArgumentParser() ap.add_argument("--data-csv", required=True) ap.add_argument("--audio-col", default="path") ap.add_argument("--out", default="data/drift_reference/train_features.json") args = ap.parse_args() df = pd.read_csv(args.data_csv) out = {"duration_sec": [], "rms_energy": [], "spectral_centroid": [], "pitch_estimate": []} for path in df[args.audio_col]: try: signal, sr = decode_audio_to_tensor(path) signal = to_mono_16k(signal, sr) features = validate_and_extract_features( signal, sample_rate=16000, min_duration_sec=0.0, max_duration_sec=1e9, rms_silence_floor=0.0 ) except Exception as e: # noqa: BLE001 print(f"Skipping {path}: {e}") continue out["duration_sec"].append(features.duration_sec) out["rms_energy"].append(features.rms_energy) out["spectral_centroid"].append(features.spectral_centroid) out["pitch_estimate"].append(features.pitch_estimate) with open(args.out, "w") as f: json.dump(out, f) print(f"Wrote reference distribution ({len(df)} clips) to {args.out}") if __name__ == "__main__": main()