01 / FOCUS TOOLS
RSVP SPEED READER
No text to display
Insert text
SPACE Play / Pause ESC ExitZEN MODE ACTIVE
02 / FOCUS TOOLS
AUDIO TRANSCRIBER
OpenAI API key
Stored only in this browser's local storage on this machine. Uncheck and hit Forget on a shared computer.
Settings
Model guide — price per minute & best use
| Model | USD / minute | Best use |
|---|---|---|
| gpt-transcribe | $0.0045 | Recommended for most recordings. Lectures, voice notes, and ordinary transcription. |
| gpt-4o-transcribe-diarize | ≈$0.006* | Recommended for multiple speakers. Interviews and meetings with speaker labels and timestamps. |
| gpt-4o-mini-transcribe | ≈$0.003 | Lowest-cost paid choice. Everyday audio when budget matters most. |
| gpt-4o-transcribe | ≈$0.006 | Established GPT-4o transcription; useful for existing workflows or comparing results. |
| whisper-1 | $0.006 | Legacy Whisper API compatibility. This app saves plain text. |
| Whisper Base · on device | $0 · no API key | Recommended free choice when your device can handle it. Favors accuracy over Tiny. |
| Whisper Tiny · on device | $0 · no API key | Lighter model for phones or when Base is too slow or runs out of memory. |
Paid models require your API key and bill your OpenAI API account. ≈ means estimated: token-based charges vary with the recording and output. *Diarize's estimate uses its published token rates, which match GPT-4o Transcribe. Local models have no API fee; downloads may use mobile data. Prices checked September 6, 2026. API pricing · Diarize rates · Model guidance
Speaker labels and timestamps are included in your text download. Vocabulary hints are unavailable for this model. For large files split into separate uploads, speaker labels restart in each part: Part 1 Speaker A may be a different person from Part 2 Speaker A.
Files under 24 MB are sent as-is. Bigger ones are decoded here in the browser, downmixed to mono and split into chunks that fit OpenAI's 25 MB upload limit, then stitched back together.
Free. No API key. Your audio stays on this device.
The first transcription downloads the selected model from Hugging Face; use Wi-Fi if possible.
Models are cached when your browser allows it, but may need downloading again if the cache is cleared.
Base favors accuracy; Tiny uses less memory. Smaller models can miss names and technical terms.
Keep this page open and your phone awake. Long recordings may be slow or exceed your browser's memory; try a shorter file if needed.
Local transcription uses automatic short sections, one at a time. Vocabulary hints and the two chunk settings apply only to paid models.
Drop audio or video files here
or click to browse — mp3, m4a, wav, mp4, mov, ogg, flac, webm · multiple files fine
Free models process audio on this device. Paid models send audio directly to
api.openai.com and bill your OpenAI API account.