Tools › Audio tools › Audio Transcription
Transcribe an audio file or a live recording using an AI speech-recognition model (Whisper). Recognition runs entirely in your browser — your audio is never sent anywhere.
Your audio is never sent to a server (the AI model itself is downloaded from a third-party site the first time you use this tool)
How to use
- Choose an audio file (mp3, wav, m4a, etc.), or record live with "Record with microphone".
- Pick accuracy and language, then press "Start transcription". The AI model downloads on first use only.
- Once done, switch the output format (plain text / timestamped / SRT / VTT / Markdown). Use SRT or VTT for video subtitles.
- Copy the result, or save it as a file.
How AI transcription works without uploading your audio
This tool runs Whisper — an open-source speech-recognition AI released by OpenAI — directly inside your browser using WebAssembly. Most transcription services upload your audio to a server for processing; here, your audio data never leaves your device. That makes it a safe choice for meeting recordings or personal voice memos you'd rather not hand to a third party.
The first time you use it, the AI model (tens of megabytes) is downloaded from a model-hosting site (Hugging Face). This is a one-way download, comparable to installing an app — nothing is sent from your side. The model is cached in your browser, so later visits start instantly.
Accuracy and speed guide
| Setting | Model size | Best for |
|---|
| Faster | ~40MB | Quick notes, rough drafts, testing |
| More accurate | ~80MB | Meetings and interviews (recommended) |
"More accurate" is recommended for real conversations. Both settings do better with quiet environments and clear speech; background noise, overlapping speakers, and heavy jargon reduce accuracy, so always review the result before relying on it. Audio longer than 20 minutes is demanding on phones — a PC is recommended for long recordings.
FAQ
Is my audio uploaded anywhere?
No. All recognition runs inside your browser. Only the AI model itself is downloaded from a distribution site on first use — that's a one-way download, and neither your audio nor the transcription result is ever sent anywhere.
What audio formats are supported?
Any format your browser can play — mp3, wav, m4a, aac, ogg, webm and more. Voice memos and meeting-recording app files usually work as-is.
It's slow or stalls partway through.
Since the AI runs inside your browser, speed depends on your device. Try the "Faster" model, close other tabs, use a PC, or split long audio into shorter pieces.
Does it work offline?
Often yes, once the model has been downloaded and is cached in your browser (this depends on your browser's settings and cache state).
Will the browser show a "page not responding" warning during processing?
No. The heavy transcription work runs in a background thread (a Web Worker) separate from the page itself, so the display and elapsed-time counter keep updating throughout, and the browser never prompts you to wait or exit.
How should I choose between SRT and VTT?
Both are video subtitle formats. SRT is the standard for YouTube subtitle uploads and editors like Premiere or DaVinci Resolve. VTT is the web-standard format used with the HTML5 video tag, for embedding subtitled video on a website. If unsure, SRT is a safe default.
Can I change the output format without re-running the transcription?
Yes. Once transcription is complete, switching the format dropdown re-renders instantly from the stored result — no need to run the AI again.