WhisperJAV

A local subtitle generator for Japanese adult videos that uses Whisper and Qwen3-ASR, with local translation through Ollama and no cloud media uploads.

WhisperJAV generates subtitles for Japanese adult videos on your own computer. It targets difficult audio with quiet dialogue, loud vocalisations and long stretches that can cause speech recognition models to invent or repeat words.

Speech recognition models: Choose Whisper, Qwen3-ASR, anime-whisper or Kotoba, plus GGUF models such as Japanese Cohere Transcribe and NVIDIA Parakeet.
Speech detection and cleanup: Voice activity detection, speech enhancement and text filters help separate dialogue from noise and reduce false subtitles.
Subtitle timing: Word alignment and sentence splitting place captions around spoken dialogue, including for models that don't produce their own timestamps.
Batch and ensemble processing: Process multiple files and run two recognition passes with different models through a graphical interface or command-line tools.
Local translation: Translate subtitles with local models through Ollama; external providers such as DeepSeek send translation requests to a cloud service.

WhisperJAV is free and open source under the MIT license. It runs on Windows, Apple Silicon Macs with macOS 14 or newer, and Linux. Media stays on your machine. Installation and initial model downloads need internet access.

CPU processing works but is slower; the stated minimum is 8 GB RAM. NVIDIA acceleration supports GTX 16 or RTX 20 cards and newer. Mac and Linux execution remain unverified by the project, and Apple GPU support hasn't been tested across all pipelines. GGUF recognition pipelines require an AVX2-capable processor.