WhisperJAV

Why we included this project

WhisperJAV tackles a specific, nasty problem: making speech-to-text work on Japanese adult video, where generic ASR models stumble badly. The audio is long and noisy, full of whispers, gasps, and theatrical vocalizations that fool standard Whisper into transcribing nonsense. The project bundles several backends, including Faster-Whisper, OpenAI Whisper, Qwen3-ASR, and anime-tuned variants, behind a common pipeline that adds scene segmentation, voice-activity detection, and hallucination filtering before emitting timestamped subtitles. It ships both a GUI and a CLI, and the decoupled ChronosJAV pipeline accepts any model that converts audio to text. If you have fought vanilla Whisper over Japanese dialogue, this is concrete engineering around the failure modes, not a thin wrapper.