How to use
- Drop a recording or video into the box.
- Choose accuracy and language, then click “Start transcribing”. The first run downloads the AI model.
- The transcript appears below as it goes. Edit it, then download as text or SRT subtitles.
How it works
It runs OpenAI's open speech recognition model Whisper (MIT license) in your browser, splitting the audio into roughly 30-second pieces at quiet points. Accurate uses Whisper large-v3-turbo; Normal uses Whisper small.
Your recording is never uploaded. Only the AI model is downloaded from Hugging Face the first time and kept in your browser.
Speed and accuracy
With graphics acceleration (WebGPU) in recent Chrome or Edge it is often much faster than real time. Without it, Normal runs on the CPU and may take longer than the recording. Overlapping speakers and loud noise reduce accuracy; double-check names and technical terms.
FAQ
Is it free? Any time limit?
Free with no usage limits. Each file can be up to 2 hours.
Is my recording uploaded?
No. Audio is processed only in your browser. Only the AI model is downloaded from Hugging Face the first time.
How big is the model download?
Normal is about 220MB and Accurate about 560MB. It is cached, so later runs start right away.
What is an SRT file for?
Upload it to YouTube Studio subtitles or editors like Premiere, DaVinci Resolve or CapCut to add timed captions.
Does it identify speakers?
Not yet. It writes what was said in time order.