Local Windows utility
Transcriber
Feed in audio or video and get a timestamped transcript out — as subtitles, plain text or structured JSON. Spark Tray can detect the language for you or take your lead, translate to English, and let you trade speed for accuracy with a model-size picker.
Windows 10/11 (x64) · Free · No account
Transcripts save to your Transcripts folder.

What it does
Everything Transcriber handles
A focused set of jobs, each done well — and nothing that phones home.
SRT, VTT, TXT or JSON
Export ready-to-use subtitles, a clean text file, or structured JSON — pick whichever your workflow needs.
Segment or word timestamps
Choose broad segment timings for subtitles or fine word-level timestamps when you need to line things up precisely.
Detect or set the language
Let Spark Tray auto-detect what's being spoken, or pin the language yourself when you already know it.
Translate to English
Transcribe in the original language, or have Spark Tray translate the speech to English as it goes.
Model-size picker
Trade speed for accuracy with a model picker. It starts on ggml-base.en — a friendly balance for everyday audio.
All on your machine
whisper.cpp runs locally, so even long or sensitive recordings never leave your PC.
How it works
From link to done in a few clicks
- 1
Add audio or video
Drop in a recording in almost any format — ffmpeg handles the decoding for you behind the scenes.
- 2
Set language & model
Auto-detect the language or set it, decide whether to translate to English, and pick a model size for the speed you want.
- 3
Transcribe locally
whisper.cpp turns the speech into text with your chosen timestamp detail, right on your machine.
- 4
Export your transcript
Save as SRT, VTT, TXT or JSON to your Transcripts folder, ready for captions, notes or further processing.
Under the hood
The engines, named plainly
Everything the Transcriber needs runs locally — here's the stack.
- Speech engine
- whisper.cpp
- Media decoding
- ffmpeg
- Default model
- ggml-base.en (swap sizes for speed vs. accuracy)
- Formats
- SRT, VTT, TXT, JSON
- Timestamps
- Segment-level or word-level
- Languages
- Auto-detect or manual, with translate-to-English
- Output folder
- Transcripts
- Platform
- Windows 10/11 (x64) — runs fully on your machine
whisper.cpp is a lean, local build of OpenAI's Whisper speech model — no cloud API, no account. The default ggml-base.en model downloads on first use; larger models trade a bit of speed for extra accuracy.
The rest of the belt
Three more tools in the tray
Questions
Transcriber, answered
Add Transcriber to your tray
Transcriber ships inside Spark Tray with the other three tools — one small download, no account.
Windows 10/11 (x64) · Free · No account