Audio transcription
Turn audio into editable text locally
Transcribe interviews, podcasts, lectures, meetings and voice notes without sending the recording to a remote transcription service. Edit individual words and export plain or timed formats.
Quick answer
Whisper Online downloads a whisper.cpp model, processes the selected media locally, lets you correct individual timed words, and exports the result without uploading the media file.
Step by step
The practical workflow
- 1
Choose the recording
Select a WAV, MP3, M4A or another browser-readable audio file from the device.
- 2
Choose a Whisper model
Use Tiny for a lightweight first pass, Base for a balance, or Small when the device has enough memory and accuracy is the priority.
- 3
Review the transcript
Correct words inline, merge punctuation, or open the detailed list to adjust each token and its timing.
- 4
Export for the next task
Download readable TXT, subtitle-ready SRT or VTT, or JSON containing every edited token and timing.
Useful for more than captions
A timed transcript can support editing, research and publishing. Podcasters can prepare show notes, journalists can review interviews, students can search lecture material, and researchers can retain structured timing alongside corrected text.
The transcript editor keeps the words connected to their original time ranges. That makes it possible to produce both a clean text document and machine-readable timing data from the same review pass.
- Podcast drafts and episode notes
- Journalistic and documentary interviews
- Lectures, study recordings and research notes
- Meeting recordings and voice memos
- Voice-over scripts and subtitle preparation
TXT, JSON, SRT or VTT?
Choose TXT when the words matter more than precise timing. The TXT exporter can optionally include caption timestamps. Choose JSON when another application or script needs every individual token's start and end time.
SRT and VTT group timed words into caption cues. SRT is a common choice for editors and media players; VTT is intended for captions attached to HTML video on the web.
What local processing changes
Local processing removes the media-upload step. The recording remains on the device, while the browser downloads a Whisper model from Hugging Face and runs whisper.cpp through WebAssembly.
This is valuable for unpublished work and sensitive recordings, but it also means the device supplies the memory and processing power. Shorter recordings and smaller models are the most reliable starting point on limited hardware.
Why use local transcription?
FAQ
Common questions
Does the audio leave my computer?
No. The selected recording is decoded and transcribed in the browser. A Whisper model is downloaded, but the audio is not sent to a transcription server.
Can I transcribe interviews and podcasts?
Yes. Clear speech with limited background noise works best. Review names, specialist vocabulary and overlapping speech before using the result.
Can I add or retime words?
Yes. The detailed token editor lets you change text and timing, delete rows and insert new timed tokens. The complete edit can be undone.
Does it identify different speakers?
No. The current transcript contains timed words but does not perform speaker diarization or label speakers.
Related guides