Kann ich Masterchannel vibecoden?

Build a local command line audio mastering tool in Python 3.11. No web UI, no accounts, no cloud, no telemetry.

Stack: Python 3.11, matchering for reference matching, pyloudnorm for loudness measurement, soundfile and numpy for IO and DSP, ffmpeg via subprocess for format conversion and MP3 export. Use a venv and a requirements.txt. No Docker.

CLI, using argparse:
  master --input mix.wav --reference ref.wav --target-lufs -14 --true-peak -1.0 --out ./out

Behavior:
1. Decode input and reference to 44.1kHz stereo 32-bit float WAV in a temp dir using ffmpeg, accepting wav, aiff, flac and mp3 inputs.
2. Print a before report for the input: integrated LUFS, loudness range, sample peak, estimated true peak, stereo correlation, and a rough 8-band spectral balance in dB.
3. Run matchering to match the input's spectrum and dynamics to the reference.
4. Apply a final loudness stage: measure integrated LUFS with pyloudnorm, apply gain to hit --target-lufs, then apply a simple lookahead true peak limiter (5ms lookahead, 50ms release) so estimated true peak never exceeds --true-peak.
5. Print an after report in the same format as the before report, plus the total gain applied and the number of limiter gain reduction events.
6. Export three files to --out: master_24bit.wav at 44.1kHz, master_16bit.wav dithered with TPDF, and master.mp3 at 320kbps via ffmpeg.
7. Write report.json next to them with all measured values for both stages.

Also support --no-reference, which skips matchering and does loudness plus limiting only. Print a warning that this is a normalizer, not a master.

Safety: never clip. Refuse to run and exit non-zero if the input is already above -6 LUFS integrated or if its estimated true peak is above 0.0 dBTP, with a message telling the user to export a quieter mix with headroom.

Include a README with install steps, one worked example, and a short note that the reference track choice matters more than any flag. Ship three pytest tests: gain math hits the LUFS target within 0.3 dB, the limiter never exceeds the ceiling on a synthetic clipping signal, and the headroom guard exits non-zero. No stems, no batch mode, no plugin hosting, no genre presets.

Antworte mir auf Deutsch. Oberfläche, Kommentare und README auf Deutsch, alle Bezeichner wie Tabellen, Spalten, Variablen, Routen und Dateinamen bleiben Englisch.