Starstruk Logo MarkStarstruk

v0.2.22

PaddlePaddle/Paddlev0.2.22Jun 4, 2026by QuentinFuxa

AI Summary

This release focuses on core audio processing improvements, specifically normalizing Chinese language codes for transcription and fixing timestamp drift issues in Voice Activity Detection (VAD). It also optimizes CUDA memory management for performance.

Key Highlights

  • NLLW language aliases normalization (zh -> zh-CN) for Whisper Chinese language codes
  • VAD event handling improvements to preserve silence boundaries and prevent timestamp drift
  • CUDA cache cleanup optimization for lightweight per-step processing

New Features

  • NLLW language aliases normalization
  • VAD event handling improvements
  • CUDA cache cleanup optimization

Full Release Notes

## Core
- **NLLW language aliases** : Normalize Whisper Chinese language codes at the NLLW boundary (`zh` -> `zh-CN`) for transcription and translation configs
- **VAD event handling** : Preserve short silence boundaries and avoid cumulative timestamp drift in VAC/VAD chunk processing
- **CUDA cache cleanup** : Keep per-step decoder cleanup lightweight; CUDA cache release stays on coarse segment boundaries