i
The full take
Video · audio · synced
Speakeasy
A note in your own voice — captured, divided, transcribed.
Camera open · ready when you are
See and hear yourself exactly as others do.
Vinh Giang's three-step technique: film yourself for five minutes, then study the take three ways — sound only, picture only, and a transcript with every um, ah and repeat marked for elimination. Speakeasy is the ritual, automated. Each recording leaves the four artefacts below behind: the original take and the three reviews — alongside a measure of how your pitch, volume, and pace move while you speak. Press the brass.
Grant the camera and microphone to begin. Nothing leaves this tab.
00:00
Open the room
⁂
Each recording leaves four pieces behind: the full take, the audio alone, the silent picture, and the words. They appear here when you stop.
Flat or full — the difference is speed, pitch, volume, pause.
Vinh Giang's vocal toolbox. Speed conveys excitement, or — slowed — gives weight. Pitch signals emotion. Volume projects confidence (or, when whispered, intimacy). Pause makes the next word matter. The analysis below measures three of those four — pitch, volume and pace, with the fourth (pause) showing up as the gaps between them. Watch the lines for stretches where they all flatten at once: that's where your delivery sat in monotone.
A note on the
fourth artefact
Transcription runs entirely in the browser. The model is OpenAI's
Whisper compiled to ONNX and served via @xenova/transformers;
on first use the weights download once and live thereafter in your
browser cache. Desktop loads base.en (≈ 145 MB); phones
load tiny.en (≈ 39 MB) so iOS Safari doesn't run out of
memory. Nothing is sent to any server.