Fixed
Multi-audio videos could be downloaded with the wrong language track. A video whose original audio is French could come out with its English auto-dub — with the right format id in the logs.
YouTube's multi-audio format ids carry a positional suffix (251-0, 251-1). HomeTube analysed the video with a bare yt-dlp -J call, picked the id of the track it wanted, then ran the download with --format-sort res,fps,+size,br — and that sort renumbers the suffixes. The id chosen during the analysis then named a different track. The part file kept the expected name and the language tag still read eng (YouTube labels every webm audio track eng), so nothing in the logs looked wrong; only playback gave it away.
Audio tracks are now requested by their language, with the positional ids kept as a fallback:
399+ba[format_id^=251-][language=fr-FR]/399+251-1
Verified end-to-end on a reported video: even when the numbering flips between the two calls, the audio in the final file is byte-identical to the intended track.
Thanks to @Sthaagg for the report and the logs that located it, and to @jaystern for identifying --format-sort as the mechanism behind the renumbering.
What's Changed
🐛 Fixes
- ci(labeler): drop the fork checkout that failed the label check on every fork PR by @EgalitarianMonkey in #145
- fix(audio): select audio tracks by language, not by position by @EgalitarianMonkey in #146
🧹 Maintenance
- chore(release): bump version to 2.12.3 by @EgalitarianMonkey in #147
Full Changelog: v2.12.2...v2.12.3