Starstruk Logo MarkStarstruk

@biomejs/biome@2.5.12 v4.12

high risk

jianchang512/pyvideotransSep 3, 2026 → Sep 6, 20263 days between releases

What's New

Introduces VibeVoice-ASR, replaces Parakeet with Nemotron, changes LLM from re-sentence splitting to correction, and removes Whisper pre-segmentation.

Breaking Changes

  • Removed 'Whisper pre-segmentation' and 'Merge short subtitles' features.
  • Changed LLM behavior from re-sentence splitting to correction only.
  • Secondary recognition now defaults to faster-whisper instead of previous defaults.

New Features

  • Added 'Auto Detect' option to the video translation pronunciation language dropdown.
  • Added VibeVoice-ASR as a new ASR channel.
  • Display all 80 available languages in independent functions.
  • GeminiAI channel fixed to use the gemini-3.5-transcribe model.

Improvements

  • Upgraded transformers library to version 5.13.1.
  • Optimized VAD algorithm for better performance.
  • Secondary recognition uses faster-whisper (large-v3-turbo) for enhanced compatibility.
  • Optimized translation prompt.

Bug Fixes

  • Fixed Qwen-asr language setting error.
  • Fixed secondary recognition to use faster-whisper.

Upgrade Notes

Users relying on removed features like 'Whisper pre-segmentation' or 'Merge short subtitles' must adapt their workflows. LLM behavior has changed from re-sentence splitting to correction; verify settings in 'Advanced Options > General Settings > LLM Correction AI Channel'. Secondary recognition now defaults to faster-whisper; check 'Advanced Options > Speech Recognition Parameters > Secondary Recognition Model' if needed.

The Two Releases

@biomejs/biome@2.5.12

Biome CLI v2.5.12

Sep 3, 2026· 2 months ago

This release focuses on resolving complex parsing errors in Astro files and introduces a new rule for flattening nested math functions.

Full release notes
v4.12

windows-v4.12

Sep 6, 2026· 2 months ago

This release introduces the VibeVoice-ASR channel, updates internal models (Nemotron and Parakeet), and refines the LLM workflow by replacing 're-sentence splitting' with 'correction'. It also fixes Qwen-asr errors and optimizes the VAD algorithm.

Full release notes