Recording and uploads.

Record

Press the red button, or press R anywhere in the app. While you talk, the words appear as live dictation from your browser's speech recognition, and the audio is kept alongside the transcript. Press Space to pause and resume.

Live dictation is provided by your browser. In Chrome the speech recognition is run by the browser vendor, so audio may leave your device. If that matters to you, use Upload or Paste instead.

If recognition fails partway, Ramble flags the transcript as possibly partial and keeps the audio. Pick the dictation language in Settings. With no choice saved, Ramble uses your browser's language.

Upload

Choose an audio file (m4a, mp3, wav, webm, ogg or anything your browser reports as audio). It is decoded in your browser and transcribed on your device by Whisper, in a web worker, so the page stays usable. Your audio is not uploaded.

• The model is onnx-community/whisper-base, pinned to one commit. The first transcription downloads its files from Hugging Face (about 80 MB, as the upload panel states). Your browser caches them after that.
• Ramble tries WebGPU first and falls back to WASM if it fails.
• Long audio is processed in 30 second chunks with a 5 second overlap. A progress bar shows loading, decoding and transcribing.
• You can cancel. A job with no progress for 120 seconds stops with an error.

Paste

Paste or type text. The composer saves your text, mood and date in the browser as you go, so a closed tab does not lose them. Audio, title and project are not part of that draft.

Before you save

Set when the thought happened, and a mood from 1 to 5. Add a title, or leave it blank and Ramble writes one.