credits for community contributions from the original faster-whisper

[BBC-Esq]

This issue tracks contributions inspired by unmerged pull requests from the original SYSTRAN/faster-whisper repository. Thanks to everyone who put in the work — your changes deserve to ship.

---

• @Br1an67 — fix for punctuation merging with multi-character tokens (e.g. French apostrophes like j', l'). Implemented in commit bca9a56.

[BBC-Esq]

• @Purfview — simplified VAD context prepending logic to avoid unnecessary array copies. Implemented in commit cde5cbc.

[BBC-Esq]

• @erichaydel — added a language field to each Segment so multilingual transcription now reports the detected language per-segment. This is especially useful for audio with mid-stream language switches (e.g. meetings mixing English and German). Opens the door for future work on mixed-language transcription and translation workflows. Implemented in commit ece617a.

[BBC-Esq]

• @extrange — fixed language_detection_threshold type hint from Optional[float] to float across all signatures, since passing None would cause a runtime error. Implemented in commit 4ccc11e.