# What changed in whisper-s2t-reborn

> 23 changes in whisper-s2t-reborn 1 from its official release notes, upgrade guides and advisories: 6 breaking, 3 new APIs and options. Each one cited. Sources checked 2026-10-24. HTML: https://unstale.sh/pypi/whisper-s2t-reborn/

To see which of these apply to a repository, run `npx @unstale/cli init` in it.

## whisper-s2t-reborn 1

### Breaking changes

- The library now requires huggingface-hub 0.23 or newer. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- Five unused decoding settings have been dropped. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- This option, which never did anything, has been dropped. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- Start times in JSON and TSV output are written as floats, such as 29.0 rather than 29. (1.7.2) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.2)
- Without VAD, a final window under 1 s is transcribed with the preceding window instead of as a separate trailing line. (1.7.2) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.2)
- CTranslate2 4.6.3 is now the minimum supported version, so every install uses its cuDNN-free pure-CUDA convolution path. (1.7.1) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.1)

### New APIs and options

- The needed aligner model can be predetermined so only that model is downloaded, and it can be run from a full offline folder. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- Models can now be loaded fully offline from a folder, enabling use in Docling's offline mode. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- The parameter can now be adjusted instead of always using its default. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)

### Behavior and default changes

- Downloading a model no longer loads PyTorch. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)

### Fixes and other changes

- Prevents duplicate words from appearing across VAD segments. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- Prevents empty transcripts from producing hallucinated text. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- Downloads work with every precision setting, can be forced to re-download, and fail immediately when a model is unavailable offline. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- The library no longer fails to import in read-only environments. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- Short audio waves are padded during extraction and processing. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- Abnormally short transcriptions are retried. (1.8.0) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.8.0)
- Without VAD, fractional max_speech_len values no longer make windows overlap and repeat words. (1.7.2) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.2)
- With word timestamps enabled, very short audio windows no longer cause transcribe_with_vad() to abort with an IndexError. (1.7.2) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.2)
- With word timestamps enabled, very short audio windows no longer cause transcribe() to abort with an IndexError. (1.7.2) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.2)
- Files without an initial prompt are no longer contaminated by prompts from other files in the same batch, making their output identical to transcribing them alone. (1.7.1) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.1)
- Benchmark scripts now exit promptly with a clear error when the worker process dies, rather than hanging indefinitely. (1.7.1) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.1)
- 8-, 24-, and 32-bit WAV files are decoded correctly instead of being misread as 16-bit and producing plausible-looking garbage with wrong timestamps. (1.7.1) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.1)
- Word-level timestamps no longer crash on large-v3, large-v3-turbo, and distil-large models; the word aligner falls back to the main model when the default tiny aligner cannot accept its audio features. (1.7.1) — [release notes](https://github.com/BBC-Esq/WhisperS2T-reborn/releases/tag/v1.7.1)