Vocal Coach AI
Cadence · Private artificial intelligence (AI) speech & communication coach
Outcome
Personal speech coaching from your own words, with privacy you control.
- ✓Feedback tied to the exact moment in your speech
- ✓Practice drills that take about a minute
- ✓Progress tracked from one session to the next
- The challenge
- Professionals know they say "just" and "um" too much or rush when nervous, but generic speaking tips don't fix it. Real coaching is expensive, and recording yourself raises privacy worries.
- How I led it
- Product owner and designer. I defined the coaching approach (coaching, not scoring), the drill library, the privacy and retention controls and the voice-consent step, and directed AI-assisted development.
- What I delivered
- Cadence listens only while you speak, transcribes and analyzes the session, and coaches from your own words: moments to replay, specific drills and a playback of how the line could sound, all with the user controlling what is kept.
Deep dive
Open any section for detailHow it worksFrom speaking to better speaking
- 01
Listen
Captures only while you speak
- 02
Analyze
Fillers, hedging, pace, tone
- 03
Coach
Moments to replay + drills
- 04
Hear it
How you could sound
- 05
Track
Check-ins over time
Challenge and solution, side by side3 examples
- ChallengeGeneric tips don't change habits.SolutionDrills are built from the user's own sentences, e.g. re-say your opening line without "just".
- ChallengeHard to hear what better sounds like.Solution"How you sounded vs. how you could sound" playback in a choice of voices.
- ChallengeWorry about being recorded.SolutionListens only during speech, lets users choose retention or delete after analysis, and asks for consent before any voice model.
Product focus and qualityFocus, validation approach and key features
Product focus
Coaching from your own words: it listens only when you speak, points out moments to replay, and gives you a 60-second drill built from what you just said.
Quality & validation
Privacy is in the user's hands: choose how long recordings are kept, or delete them right after analysis. A voice model is only created after explicit consent. Coaching draws on a curated library of established speech techniques rather than open-ended AI advice.
Key features
- ·Listen mode that captures only while you speak
- ·Session history with moments to replay
- ·Coaching on filler words, hedging, pace, tone and articulation
- ·Drills built from your own speech, e.g. re-say an opening line without "just"
- ·"How you sounded vs. how you could sound" playback in a choice of voices
- ·Check-ins over a chosen time window, with progress from session to session
- ·Retention controls and consent before any voice model
Why it matters
Shows I can work with live audio and AI while keeping privacy and consent in the user's control.
Scope, notable components and compliance
Measurable scope
- ·Coaching library of established techniques (pace, fillers, hedging, breath, presence)
- ·Session history with check-ins over a chosen window
Notable & novel components
- ·Drills generated from the user's own speech
- ·Before/after voice playback of the user's lines
- ·Coaching grounded in a curated technique library
Compliance work
- ·User-set recording retention, including delete right after analysis
- ·Explicit consent before creating a voice model
Inside the app15 screens
Tech stack and skills
Tech stack
TanStack Start (React server-side rendering (SSR))TypeScriptTailwind / shadcnLovable Cloud (Postgres, auth, storage)Lovable AI Gateway (speech-to-text, analysis, text-to-speech)
Skills demonstrated
Speech-to-text & audio AICoaching designPrivacy by designConsent design
▤ README.md›
# Vocal Coach AI > Cadence · Private artificial intelligence (AI) speech & communication coach "Cadence": a private AI speech coach that listens while you talk, finds filler words, hedging, pace and tone, and builds short drills from your own speech, then plays back how you could sound. ## Outcome **Personal speech coaching from your own words, with privacy you control.** - Feedback tied to the exact moment in your speech - Practice drills that take about a minute - Progress tracked from one session to the next ## Problem Professionals know they say "just" and "um" too much or rush when nervous, but generic speaking tips don't fix it. Real coaching is expensive, and recording yourself raises privacy worries. ## Solution Cadence listens only while you speak, transcribes and analyzes the session, and coaches from your own words: moments to replay, specific drills and a playback of how the line could sound, all with the user controlling what is kept. | Problem | Solution | | --- | --- | | Generic tips don't change habits. | Drills are built from the user's own sentences, e.g. re-say your opening line without "just". | | Hard to hear what better sounds like. | "How you sounded vs. how you could sound" playback in a choice of voices. | | Worry about being recorded. | Listens only during speech, lets users choose retention or delete after analysis, and asks for consent before any voice model. | ## Features - Listen mode that captures only while you speak - Session history with moments to replay - Coaching on filler words, hedging, pace, tone and articulation - Drills built from your own speech, e.g. re-say an opening line without "just" - "How you sounded vs. how you could sound" playback in a choice of voices - Check-ins over a chosen time window, with progress from session to session - Retention controls and consent before any voice model ## Tech stack - TanStack Start (React server-side rendering (SSR)) - TypeScript - Tailwind / shadcn - Lovable Cloud (Postgres, auth, storage) - Lovable AI Gateway (speech-to-text, analysis, text-to-speech) ## Notable components - Drills generated from the user's own speech - Before/after voice playback of the user's lines - Coaching grounded in a curated technique library ## Compliance - User-set recording retention, including delete right after analysis - Explicit consent before creating a voice model ## Scope - Coaching library of established techniques (pace, fillers, hedging, breath, presence) - Session history with check-ins over a chosen window ## Links - Live: https://vocally-bright.lovable.app ## Author Candace Hagood · Product & Quality Leader · AI-Assisted Engineering