Private audio & video transcription
Transcription uses a speech model that runs in your browser (via a Web Worker). Your recording is not uploaded to FlashKit. The first visit downloads model weights to the browser cache — that download is expected and can take a minute on a slow connection.
Accuracy tracks recording quality: a close mic in a quiet room beats a phone at the back of a hall. Accents, overlap, and loud music will insert errors. Always proofread before you publish minutes.
How to Use
- Allow the one-time model download if the status bar asks for it.
- Select an audio or video file the browser can decode (MP3, WAV, MP4, and similar).
- Pick the spoken language when the UI offers it — the wrong language hurts accuracy.
- Start transcription, then copy or export the text. Proofread names and numbers.
Key Features
- On-device model — FlashKit does not host your media
- Audio and video containers the browser can decode
- No account or per-minute FlashKit fee
- First-run model download is cached for later sessions
Frequently Asked Questions
Is this 100% private?
FlashKit does not receive the file. The browser still downloads model files from the model CDN on first use. If you paste a remote media URL, that fetch may use a proxy — prefer uploading a local file.
What languages work?
The worker supports a wide set of languages. Quality varies. English is usually strongest; always review legal or medical text.
Why is the first run slow?
Weights download once, then inference uses your CPU/GPU. Subsequent files on the same device start faster.
Can I transcribe a two-hour meeting on a phone?
You can try, but long files on mobile often run out of memory. Split with the audio cutter first.