Waverd
AI Presentation Coach
An AI-powered presentation coaching tool that records speech, transcribes it using on-device Whisper, detects filler words, and provides per-slide analytics synchronized with Apple Keynote.
Overview
Rehearse with feedback that stays on your Mac.
Waverd helps presenters review how they spoke, not just what they said. It records a rehearsal, transcribes speech on-device, flags filler words, and connects the analysis to Keynote slides so feedback has context. Speech processing runs locally rather than through a cloud API.
Engineering decisions
- On-device Whisper Large v3 Turbo (~632MB CoreML model)
- Dual filler detection: heuristic + Silero VAD neural model
- Keynote slide synchronization via AppleScript polling
- Privacy-first: zero cloud API calls for speech processing
Built with
- Swift
- SwiftUI
- WhisperKit
- CoreML
- AVFoundation
- AppleScript
A closer look
Behind the build.
Presentation feedback needs more than a transcript.
A speaker needs to know where they hesitated, how often fillers appeared, and which slide was on screen at the time. Waverd was shaped around a rehearsal that can be reviewed slide by slide, while keeping recorded speech on the Mac.
Bring audio, words, and slides onto one timeline.
AVFoundation captures the rehearsal, WhisperKit produces on-device transcription with word timestamps, and an AppleScript monitor checks the active Keynote slide. Slide-change times divide a complete recording into per-slide sessions, so the transcript and playback retain presentation context.
Combine text and acoustic signals.
Filler analysis uses a word list and timing heuristics alongside audio events from a Silero voice-activity model. The results are merged into an annotated transcript instead of treating every pause or short word as the same kind of mistake. The speech-processing path runs locally, without a cloud transcription API.
Inside the product
From the app itself.
