How to transcribe a recording
- Drop your audio or video above. Phone voice memos, Zoom and Teams recordings, podcasts and lecture captures all work.
- Choose a model. Fast (about 40 MB) for quick drafts, Balanced (about 80 MB) for most recordings, or Accurate (about 250 MB) for difficult audio. It downloads once and is cached.
- Pick the language or leave it on auto-detect.
- Read and edit the transcript as it appears. Each segment has a timestamp; click ▶ to hear that part again and correct anything that was misheard.
- Download as TXT, TXT with timestamps, Word (DOCX), SRT or VTT.
Getting the most accurate transcript
- Record close to the speaker. A phone on the table near the person talking beats a laptop microphone across the room.
- Reduce background noise. Air conditioners, music and cafés hurt accuracy more than accents do.
- One person at a time. Overlapping speech is the hardest thing for any recognizer.
- Use the Accurate model for important recordings, strong accents or technical vocabulary.
- Fix names with Find and replace. If a person’s name or a product is misheard, replace it everywhere in one go.
Who uses private transcription
Journalists and researchers transcribing interviews that contain sources’ details or participants’ personal data, which research ethics often say shouldn’t be uploaded to third-party services. Students turning lectures into study notes. Lawyers, doctors and therapists making notes from dictation. Podcasters creating show notes and searchable transcripts. Teams keeping a written record of meetings without sending recordings to another company.
How it works
The tool uses OpenAI’s open-source Whisper speech recognition model, converted to run in the browser with transformers.js. If your browser supports WebGPU, the model runs on your graphics chip; otherwise it falls back to WebAssembly on your CPU. Long recordings are split into five-minute chunks at quiet moments, so you can see progress, and memory use stays low.
Limits
Accuracy depends on audio quality, and like every speech recognizer, Whisper occasionally invents or repeats words during long silences or music. Always review important transcripts before relying on them. Speaker labels aren’t automatic. Very long recordings (several hours) can be slow on older devices — you can split them and transcribe in parts. Need captions on a video instead? Use the subtitle generator.
Frequently asked questions
Is the transcription really free?
Yes, with no minute limits. The AI model runs on your own device, so there's no per-minute server cost to pass on.
Is it safe for confidential recordings?
Your audio is processed inside your browser and is never uploaded. That makes it suitable for interviews, medical or legal notes, and internal meetings — but remember that anyone using your device could still access files you download.
Which file types can I transcribe?
Most audio and video formats that your browser can play, including MP3, WAV, M4A, AAC, OGG, OPUS, FLAC, MP4, MOV and WEBM. iPhone voice memos (M4A) work too.
Does it identify different speakers?
Not automatically. The transcript is split into timestamped segments, and you can type speaker names in the editor. Find and replace helps if you label speakers as you go.
Can I get the transcript in Word?
Yes. Download it as DOCX with timestamps, plain TXT with or without timestamps, or as SRT/VTT caption files.
How long does it take?
With WebGPU on a modern laptop, often faster than real time. On older devices without WebGPU, expect roughly the length of the recording or more with the Balanced model. The Fast model is quicker.
Does it work offline?
After the model has been downloaded once, yes. The model is cached by your browser, so later transcriptions can run without an internet connection.
Can it translate?
This tool transcribes in the spoken language. Choose the language manually if auto-detect picks the wrong one for a short or noisy clip.