Quick Answer
A voice note structurer turns unscripted speech — recordings rambling through several topics — into a note you can actually use: a title, a summary paragraph, key points, action items, and a cleaned transcript with filler words removed. Record in this tool or drop in an audio file; Whisper transcribes it on your device, and the structuring rules run locally in your browser. It is free, needs no sign-up, and your audio is never uploaded.
How It Works
Provide your audio: press Record and speak (a live timer and level meter confirm capture), or drop an audio/video file — the recording never leaves this device
Pick a Whisper model: tiny.en (~40 MB, fastest, English only) or base (~80 MB, multilingual) — a one-time download from the CDN, cached by your browser afterwards
Transcribe on this device: Whisper runs in the tab via WebGPU or WebAssembly and returns a timestamped transcript with chunk boundaries
Review the structured note: the local rule layer splits paragraphs (pause or topic shift), strips standalone fillers, pulls action items, TF-ranks key points, and titles the note
Edit and export: fix anything in the editable markdown — copies and downloads use your edits — then copy, download .md/.txt, or re-run the rules on the original transcript
Key Facts
- Record with your microphone (Opus/WebM with MP4 fallback) or drop an audio/video file
- Whisper tiny.en (~40 MB) or base (~80 MB) transcribes fully on this device — WebGPU where available, WebAssembly otherwise
- Timestamped segments drive paragraphing: splits on pauses over 0.8 s or a drop in topic-word overlap
- Conservative filler cleanup removes standalone um, uh, er, hmm, and hedging like — and reports the count
- Action items are pulled from commitment phrases like "need to", "remember to", and "make sure"
- Key points and the summary come from term-frequency scoring of the cleaned transcript
- Edit the markdown note, copy it, or download it as .md or .txt
- Re-run the structure rules on the original transcript at any time
- Free, no account, nothing uploaded — the only download is the open-source model file, cached after first use
Frequently Asked Questions
How does the structuring work?
The rule layer runs locally on the timestamped transcript. Paragraphs split where the pause between segments exceeds 0.8 seconds or where the overlap between the top words before and after a boundary drops below 0.15 (a topic change). Standalone fillers — um, uh, uhh, er, hmm, and hedging like — are removed only when they sit between commas or at a sentence edge, so "I like pizza" keeps its verb. Sentences containing commitment phrases (need to, have to, remember to, don’t forget, should, make sure, action item) become action items, term-frequency scoring picks the top three non-action sentences as key points, and the first short sentence becomes the title.
Is this voice note tool free?
Completely free and unlimited — no account, no trial, no watermarks. Subscription notetakers process your audio in the cloud: Otter.ai Pro is listed at $16.99/user/month on monthly billing. This tool runs the entire pipeline — recording, transcription, structuring — inside your browser at no cost, and a synthetic sample transcript demos the structure rules without any model download.
Is my audio private?
Yes. Audio is transcribed on this device; nothing is uploaded. The AI model runs in your browser tab via WebAssembly/WebGPU, and the only network traffic is the one-time download of the open-source Whisper model file itself from a CDN (~40 MB for tiny.en, ~80 MB for base). Once cached by your browser, the tool keeps working offline.
How accurate is the transcription?
It uses OpenAI Whisper, but the browser-sized models trade some accuracy for speed and privacy: tiny.en (~40 MB) is fastest and English-only; base (~80 MB) handles accents and multiple languages better. Clear speech close to the microphone works well; heavy background noise, crosstalk, and unusual names or jargon will need fixes. The structured note is fully editable, and the rules can be re-run on the original transcript at any time. For critical wording, expect to proofread.
What audio and video formats can I use?
Anything your browser can decode: MP3, WAV, M4A, AAC, OGG, FLAC audio, and most MP4/WebM/MOV videos in Chrome, which extracts the audio track locally. Other browsers are pickier with video containers — if decoding fails, the tool says so and suggests an audio export or a different format. There is no fixed length limit; longer recordings take proportionally longer to transcribe (30-second chunks are processed with overlap).
What do I get out, and can I edit it?
A markdown document with a # title, a one-to-two sentence summary, ## Key points, ## Action items, and ## Transcript (cleaned and paragraph-split), plus word count, duration, and filler-removal stats. The note is a live editable text box — copy and download always use your edited version. A Re-run button rebuilds the note from the original transcript if you want the rules applied fresh.