Speak

See what you say — and how you're saying it.

website · 2025 · no public repo

Web Speech API · Firebase · JavaScript · Canvas

speak-1337.web.app

Most speech-coaching tools need a recording, a subscription, or both. Speak runs in the browser, watches your delivery in real time, and flags monotone or rushed segments before the moment slips by.

Speak app showing live FLAT/SPEED/PITCH/VOLUME/PAUSE tracks across a recorded session, with a fillers panel and per-segment transcript — Practice mode. Live tracks for everything that betrays you.

Session complete modal showing 'A touch slow' verdict with 77 words/min, 0:20 duration, 3.0 fillers/min, suggesting a 30-second timer for the next round — Session done. Verdict, numbers, what to fix next.

Send session to AI modal showing provider options (Claude, ChatGPT, Gemini, Perplexity), markdown vs JSON format, and a prefilled prompt with structured speech metrics — Hand-off to AI. Pick a provider, format, and prompt is prefilled.

Claude analysing the speech session — diagnosing pacing as the primary issue, calling out the headline finding (silence not speech rate), and noting limited pitch variety — AI feedback. Honest critique, beyond what the numbers show.

1 / 4

Speak is a real-time speech coach in the browser. As you talk, it tracks pitch, speed, volume, pauses, and filler words — and visualises the lot as a live mini-dashboard. You can replay segments, see where you went monotone or rushed, and watch a "FLAT" bar fill as your delivery becomes more varied.

Built for impromptu practice, presentation rehearsal, and anyone who wants honest feedback without recording themselves on Zoom.

Highlights:

Live pitch / speed / volume / pause tracking
ASR transcript with per-segment WPM and pitch range
Flags for monotone, fast, slow, filler-heavy delivery
Daily streaks for practice habits
Optional "send to AI" for deeper feedback