WhisperJAV is an open-source speech-to-text subtitle generator specialized for Japanese audio and video. It combines multiple automatic speech recognition models — including Whisper and Qwen3-ASR — with voice activity detection to produce accurate subtitle files even from noisy recordings. The project has earned more than 2,300 stars on GitHub and was still actively updated in October 2026. Note that the listing link points to the GitHub repository, since the project’s download page is a release listing rather than a product page.
General-purpose transcription tools often fall apart on difficult Japanese media audio, where background music, overlapping effects, and soft-spoken dialogue cause hallucinations and missed lines. WhisperJAV is engineered specifically for that scenario: a TEN-VAD voice activity detector isolates actual speech, the audio can be enhanced before recognition, and several ASR backends — Whisper variants, Qwen3-ASR, and Zipformer among them — can be selected or combined depending on the material. Local LLM post-processing helps clean up the recognized text, and the output is a ready-to-use subtitle file. Packaged releases make it usable without assembling the pipeline yourself. Configuration options let users trade recognition speed against accuracy depending on how much media they need to process and how difficult the source audio is.
WhisperJAV is free and open source under the MIT license. Everything runs locally, so there are no per-minute transcription fees; you only supply the hardware.
WhisperJAV is for fansubbers, translators, archivists, and researchers who need dependable subtitles from Japanese media that defeats general transcription services. It is built for people who process a lot of difficult audio and need a specialized, fully local pipeline.