WhisperDesk
Local, GPU-accelerated transcription for Windows. Transcribe files and live audio, identify who said what, and export to every format you need — nothing ever leaves your PC.
Built on whisper.cpp (Vulkan: AMD, NVIDIA and Intel GPUs) and sherpa-onnx speaker models. Ideal for meetings, calls, interviews, podcasts and streamers who want live captions.
- Files & folders — drag & drop, queue, Whisper models Tiny → Large-v3 (Turbo recommended)
- Speaker identification — every line tagged with time and speaker, auto speaker count
- Voice library — remember named voices and recognise them automatically
- Live transcription — mic, Voicemeeter or PC loopback, with live speaker tagging
- Recordings library — folders, tags, notes, metadata and full-text search
- Audited edits — every correction logged with reason, date and user
- Offline call summary — key points, actions, decisions and questions
- Exports — TXT, SRT, VTT, JSON, Word, PDF and burned-in subtitles
- Share — self-contained web player or .wdsession package
- OBS caption overlay — live captions for your stream via a browser source
- Add-ons — live text file, webhook, and a developer SDK
- Themes — darkcipher0x terminal theme in dark, light or system mode
> What's new in 1.4 v1.4.2
- New install folder (1.4.2) — installs to
C:\Program Files\DarkCipher Systems\WhisperDesk. Upgrading moves it from the oldC:\Program Files\WhisperDeskautomatically; your data is untouched. - Open source (1.4.1) — now MIT-licensed, shown on the launch screen, in Help → About and in the installer.
- Stability on long jobs — the window never blocks; new Performance mode (Light / Balanced / Maximum) keeps your PC usable while it works, and the PC stays awake during jobs.
- Accuracy — Skip silence sends only speech to Whisper; a hallucination filter drops invented "Thank you." lines and repeat loops; smarter speaker detection merges look-alikes.
- Custom models — import any whisper.cpp GGML model (fine-tuned, distilled, quantised, language-specific).
- Live feed window — a clean, chat-style read-only feed with speaker colours; snap, full screen, always on top.
- OBS overlays — three looks: darkcipher0x terminal, stacked chat and classic subtitles.
- Reports on many recordings — combined or separate reports with participation, keyword hits, edit log, SHA-256 integrity and an optional local-AI consolidated report.
- Voice & audio analysis — voice profiles, voice comparison and audio integrity checks (metadata, clipping, digital silence, mains hum / ENF).
- Encrypted backups — library, voices, settings and add-ons, optionally AES-256-GCM encrypted, verified on restore.
- Summary transparency — every AI summary records which model wrote it, where it ran and how long it took.
- Library, sharing & scheduling — folder tree with drag and drop, re-transcribe, overnight scheduling, and the Windows share sheet.
- Installer — upgrade / repair / remove detection, optional FFmpeg and Ollama via winget, and per-item data removal on uninstall.
Full details in the User & Admin Guide.