Can't find your answer? Contact support — we respond within 1 business day.
ScribeZapp is a Windows desktop application that transcribes audio and video files entirely on your own computer — no internet connection required during transcription. It uses OpenAI's Whisper model (via faster-whisper) for transcription, and optionally pyannote.audio for speaker diarization.
Never. Your audio and video files are processed entirely on your Windows PC. The AI models run locally — no audio, no transcripts, and no personal data are transmitted anywhere during transcription. ScribeZapp only contacts the internet during initial setup (to download models and activate your license) and once per month to verify your subscription is still active.
ScribeZapp runs on Windows 10 and Windows 11 (64-bit only). A macOS version is not currently planned. The application is a native Windows desktop app built with Electron and a Python/FastAPI backend, compiled with Nuitka.
No. ScribeZapp is designed to work on any modern 64-bit Intel or AMD CPU. The faster-whisper engine uses CTranslate2 INT8 quantisation specifically to make GPU-quality transcription practical on CPU hardware. If you have a CUDA-capable NVIDIA GPU, ScribeZapp can optionally use it to speed up transcription — but it is not required.
ScribeZapp supports MP3, MP4, WAV, M4A, FLAC, OGG, MKV, MOV, AVI, and WebM. Audio is extracted from video files automatically using the bundled FFmpeg library. If a format can be read by FFmpeg, ScribeZapp can transcribe it.
ScribeZapp supports 99+ languages via OpenAI Whisper's multilingual model, including English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Polish, Japanese, Chinese, Korean, Hindi, Arabic, Turkish, Ukrainian, and many more. Language detection is automatic — if you don't specify a language, Whisper will detect it from the audio.
ScribeZapp exports transcripts as plain text (.txt), SRT subtitles (.srt), VTT subtitles (.vtt), and structured JSON (.json). Pro adds speaker labels in all export formats, word-level timestamps in SRT/JSON output, plus DOCX and anonymized PDF export.
Accuracy depends on audio quality, accent, background noise, and the Whisper model size. On clean audio with a clear speaker, ScribeZapp produces accuracy comparable to leading cloud services. Accuracy is generally better on clear, well-recorded audio, and may be lower on noisy recordings, heavy accents, or highly technical vocabulary — Pro's custom vocabulary feature can help with the latter.
Transcription speed depends on your CPU, the Whisper model size, and audio length. As a rough guide, on a modern CPU (e.g. Intel Core i7), faster-whisper processes audio at approximately 2–4× real-time — meaning a 30-minute recording takes about 8–15 minutes. GPU acceleration, if available, can increase this to 10–30× real-time.
Speaker diarization is the process of identifying who spoke when in a recording. ScribeZapp Pro adds speaker labels to your transcript — e.g. [Speaker 1, 00:00:12] and [Speaker 2, 00:00:18] — so you can tell different voices apart without manually reviewing the audio.
No. ScribeZapp bundles the pyannote.audio community models directly in the installer. You do not need a Hugging Face account, an API token, or any internet connection to run diarization. Everything is stored locally on your PC and runs entirely offline.
ScribeZapp can automatically detect and label multiple speakers. The pyannote.audio model works best with 2–6 speakers per recording. Performance may degrade on recordings with many overlapping voices, very short speaking turns, or significant background noise. For best results, set the expected number of speakers manually if you know it in advance.
The current version labels speakers as Speaker 1, Speaker 2, etc. Speaker renaming (replacing generic labels with real names) is on the roadmap for a future update. For now, you can use Find & Replace in your word processor after exporting the transcript.
ScribeZapp Pro scans your transcript and automatically detects personally identifying information — names, email addresses, phone numbers, and similar identifiers — replacing them with placeholders. Detection runs entirely offline on your machine.
Yes. ScribeZapp keeps a local anonymization map on your machine so you can deanonymize a transcript at any time, restoring the original names and details. This map never leaves your device.
No. ScribeZapp does not operate an AI backend and does not see your transcript. AI Summary anonymizes your transcript locally, then connects to the AI assistant you already use directly from your machine. The connection and the AI provider are yours — ScribeZapp has no visibility into that exchange.
AI Summary is designed to work with the AI assistant you already use. Exact provider support is being finalized — check the Download page or release notes for the latest details.
ScribeZapp Core requires no trial, no account, and no credit card — just download the installer and start transcribing. It's free to use indefinitely.
ScribeZapp Pro includes a 30-day free trial. After subscribing through Gumroad, the installer downloads automatically. On first launch, Pro enters trial mode with full Pro functionality for 30 days. After that, an active subscription is required for Pro features. Your transcripts and settings are always preserved.
Each license key activates on one PC at a time. If you replace your computer, email support@scribezapp.com with your Gumroad order number and we'll reset your activation so you can activate on the new machine.
Log in to your Gumroad account and cancel the subscription from the Library tab. After cancellation, ScribeZapp will continue to work until the end of your current billing period, then enter limited mode.
When your annual subscription expires, ScribeZapp enters a restricted mode — you can still view past transcripts, but new transcription requires an active subscription. Renewing restores full functionality immediately. Your downloaded AI models are kept on your PC, so reinstallation is not required.
Operating System: Windows 10 or 11 (64-bit)
CPU: Any modern 64-bit Intel or AMD processor
RAM: 8 GB minimum; 16 GB recommended for larger files and diarization
Disk: ~2–4 GB for models depending on plan
Internet: Required during installation only. Not required for transcription.
Yes, the installer is safe and code-signed. Windows SmartScreen displays a warning for new publishers until enough installs are recorded. Click "More info" then "Run anyway" to proceed. The installer is digitally signed by Edgelytix Technologies Incorporated.
The AI models are large files (1–3 GB depending on plan) and download during installation. On a slower connection this can take 10–30 minutes. As long as you see download progress, it's working. Do not close the installer mid-download. If the download is interrupted, re-run the installer and it will resume.
We reply to every message within 1 business day.