SpeechCoach
Grammarly for talks. Rehearse out loud and get a score, rewrites and a live voice coach.
- Part no.
- A93-10
- Status
- working
- Platform
- Web
- Year
- 2026
- Stack
- NodeVanilla JSWebRTCOpenAI RealtimeStructured outputs
You only find out a talk had too many "ums" and rushed endings once you're up there giving it.
A cheap, private rehearsal partner. The feedback is grounded in measured pace and filler counts, not vibes, and costs are clear per mode.
Subscription, ≈ €12/mo for about 2 hours of rehearsal, or seats sold to sales and leadership teams.
- Rehearsal minutes included per month120
- ×Live voice coach cost$0.05/min
- =Worst-case cost per user≈ $6/mo
- −Price€12/mo
- =Gross margin at full usage≈ 50%
- Live voice coaching
- $0.05/min
- Reports (transcribe + score)
- cents per rehearsal
- Hosting
- one small Node server
This is the one project where costs grow linearly with use, so minutes need to be metered and capped. Selling B2B to teams makes the margin safer, because most seats are used less than the cap.
Viable if usage is metered. Sell to teams, not to individuals.
- 01SpeakIn the browser. Audio timing is analysed locally from energy levels.
- 02TranscribeSpeech-to-text with timestamps.
- 03ScoreStructured JSON report: six sub-scores, quoted lines and drills.
- 04Live coachOptional full-duplex voice over WebRTC, either silent or chiming in.
Grounded feedback
The live coach’s tips are tied to measured filler counts and pace, so it can’t make things up about your delivery.
Zero dependencies
Node 22 standard library and vanilla JS, plus a mock mode for testing without spending API credits.