Repository navigation
Local Whisper speech-to-text with a desktop interface. Audio or video in,
accurate timed transcripts and subtitles out, entirely on your own machine.
Download
| platform | file |
|---|---|
| Windows | HyperWhisper-Setup-1.0.0.exe |
| Linux | HyperWhisper-1.0.0-linux-x86_64.tar.xz |
Python is not required — the interpreter and every dependency, including
ffmpeg, are inside the bundle.
A .sha256 accompanies each file if you want to verify the download.
What it does
- Transcribes 30 audio and video formats; video has its audio extracted for you.
- Uses an NVIDIA GPU automatically when there is one, CPU otherwise.
- Exports SRT, VTT, plain text, and a Premiere Pro-friendly SRT variant, with
long lines split into readable cues. - Batch queue for working through a folder unattended.
- Built-in viewer with search and editing, so a misheard name can be corrected
once rather than lived with. - Optional speaker labels, using a model that needs no account or token.
- Translates any supported language into English in a single pass.
Nothing is uploaded, there is no API key, and it keeps working offline once a
model has been fetched.
First run
Pick a media file, pick an output folder, press Start. The first run downloads
the default model (about 1.6 GB) once; later runs start from cache.
The README covers every setting, how to choose a model, and troubleshooting.
Notes
- The first transcription downloads a model, so it takes noticeably longer
than subsequent ones. - Speaker diarisation is an optional extra and pulls in PyTorch.
- Settings, presets and logs live in a per-user data directory, not in the
install folder.