Scribe

Scribe is a fast, offline subtitle generator and transcription tool powered by the Whisper large-v3 AI model.

Drop a file. Get subtitles. That's it.

Scribe takes any audio or video file and generates timed subtitles (.srt) or plain text transcripts (.txt). Everything runs locally on your machine.

No uploads. No accounts. No subscriptions.

On first launch, Scribe will prompt you to download a Whisper model (~3 GB for large-v3). After that, everything runs fully offline. No setup, no API keys.

Cloud transcription tools charge per minute. Scribe runs locally and costs once.

Features

• Drag & drop any audio/video file (mp3, wav, mp4, mkv, webm, flac, etc.)
• Timed .srt subtitles or .txt transcripts
• 99 languages with auto-detection
• GPU accelerated via Vulkan (works on CPU too, just slower)
• One-time model download on first launch (~3 GB), then fully offline

Who's this for?

Content creators — Generate subtitles for YouTube, TikTok, and video platforms without paying per minute.

Podcasters — Turn episodes into transcripts instantly.

Students — Transcribe lectures or recordings for easier studying.

Translators — Create a first-pass transcript to clean up.

Privacy-conscious users — Your audio never gets uploaded to a server.

Why Scribe?

Most transcription tools are cloud services. That means uploading files, waiting for processing, paying per minute, and trusting someone else with your audio.

Scribe works differently:

• No uploads — everything stays on your machine
• No subscriptions — pay once, use forever
• No waiting — processing starts immediately, on your hardware
• No internet needed — works fully offline

Language Support

Whisper supports 99 languages. How well it works depends on the language:

Works great: English, Chinese, German, Spanish, French, Japanese, Portuguese, Russian, Korean, Italian, Dutch, Polish, Swedish, Turkish, Arabic — basically anything with lots of training data.

Works well: Most European, East Asian, and South Asian languages.

Hit or miss: Rarer languages may have reduced accuracy. We're looking into ways to improve this.

Audio quality matters too. Clean recordings (podcasts, interviews, lectures) give the best results. Heavy background music or sound effects will reduce accuracy.

What's Next

• Translation — transcribe in one language, translate to another. All local, no cloud.
• Vocal isolation — strip background music/SFX for cleaner transcription
• Video tools — trim clips, burn subtitles into video
• Extension packs — specialized models for specific content
• More based on what you all ask for

Buy early, get all future updates at no extra cost

A few things to know

• This is AI transcription. It's good, but it's not perfect — no tool is. Expect occasional errors with accents, mumbling, overlapping speakers, or noisy audio.
• Some languages work much better than others.
• This is a transcription tool, not a translator (yet — that's coming in v2).

System Requirements

Minimum:

• OS: Windows 10 (64-bit)
• RAM: 8 GB
• Storage: 4 GB (includes the AI model)
• CPU: Any modern x64 processor (it'll work, just slow)
• GPU VRAM: 4 GB (if using GPU mode)

Recommended:

• RAM: 16 GB
• GPU: Any Vulkan-compatible GPU with 6+ GB VRAM (NVIDIA, AMD, Intel Arc)
• A GPU makes a huge difference in speed

Credits

Powered by OpenAI Whisper (MIT License)