HyperWhisper
Local speech-to-text app for Windows and Linux. Transcribe audio and video offline with Whisper, NVIDIA GPU acceleration or CPU, and subtitle exports.
HyperWhisper is a desktop speech-to-text app for Windows and Linux that keeps recordings on your computer. It's for people captioning videos, transcribing interviews or podcasts, or turning recordings into searchable text. The interface brings transcription, editing and subtitle export into one place.
It runs OpenAI's Whisper models through faster-whisper and CTranslate2, with NVIDIA GPU acceleration and automatic CPU fallback. Models download from Hugging Face once, then work offline from the local cache. Transcription doesn't require an API key or upload your media. The default model is large-v3-turbo; other choices include large-v3 and smaller Whisper models for different speed and accuracy needs.
A batch queue processes multiple files, while the built-in transcript viewer lets you search and correct text before export. It accepts audio and video, extracting the audio from video files automatically. Exports include SRT, VTT, plain text and an SRT variant for Adobe Premiere Pro. Long subtitle segments split into shorter cues, and sentence or word timestamps support captions and precise editing.
Optional speaker diarisation adds speaker labels to transcripts and subtitles. Its default model needs no account, though gated pyannote models require a Hugging Face account and token. The app can also translate supported languages into English and filter silent passages to reduce invented text. HyperWhisper uses a proprietary license. The source code isn't published, and the license permits personal and internal business use but prohibits redistribution and resale.
