The Hainanese chicken rice of Whisper forks

What it does

This is a Windows-heavy audio and video transcription tool built on Faster Whisper, optimized specifically to translate Japanese speech into Chinese subtitles. It wraps a custom fine-tuned model—nicknamed “ChickenRice v2”—alongside a voice-optimized VAD detector, and ships as pre-built binaries targeting specific NVIDIA CUDA and AMD RDNA architectures. If you lack a local GPU, it can fall back to Modal cloud inference.

The interesting bit

Most Whisper wrappers claim to be general-purpose; this one narrows its scope to a single language pair and bundles hardware-specific ZIP files down to the AMD gfx architecture level. The included ASMR-optimized VAD model indicates the focus is on cleanly extracting voice from challenging audio, not just running stock inference.

Key highlights

• Japanese-to-Chinese translation via the “ChickenRice v2” model, fine-tuned on 5,000 hours of audio
• Voice-optimized VAD (ASMR-onnx) for speech activity detection
• Pre-built releases for NVIDIA CUDA 11.8/12.2/12.8 and AMD RDNA1 through RDNA4 gfx families
• Modal cloud GPU fallback for machines without local graphics hardware
• Drag-and-drop batch workflow with SRT/VTT/LRC output and smart file caching

Caveats

• AMD iGPUs (gfx115x) are explicitly not yet supported
• The base ~2.2GB package excludes the Whisper model; the ~4.4GB “ChickenRice” edition includes it
• Release packaging is Windows-centric, built around batch files and pre-built executables

Verdict

Worth a look if you are doing Japanese-to-Chinese fansubbing or transcription and want a tuned model plus sensible defaults. Skip it if you need general-purpose multilingual Whisper or a polished cross-platform GUI.