BoldVoice

Problem

BoldVoice (YC S21) is an accent training app that helps non-native English speakers speak clearly and confidently. It combines video lessons from Hollywood accent coaches with real-time AI speech feedback. 5M+ downloads, 4.8 stars, featured on Forbes and TechCrunch. I downloaded the app, went through onboarding, assessment, and practice sessions to identify opportunities for improvement.

My Role

Independent product analysis — onboarding walkthrough, assessment completion, practice sessions, user review analysis, and competitive evaluation.

Process

• Went through the full onboarding: value prop screens (Hollywood coaches, AI feedback, SpeechScan), expectation-setting timeline (Week 1 → Year 1+), pricing comparison vs 1:1 coaching ($150/yr vs $150-300/hr), and a transparent 7-day trial flow with Day 5 reminder and opt-out toggle.
• Completed the 20-sentence initial assessment. Scored 90% overall. The results screen shows Top 3 Strengths (N: 94%, K: 93%, NG: 93%) and Top 3 Focus Areas (TH: 58%, Z: 84%, OH vs OW: 87%) — color-coded green/yellow/red. Sound-level highlighting in sentences shows exactly which letters were mispronounced.
• Went through the practice session: 'Challenge Words' targeting my weak sounds. Each sentence gets a circular score (87-99%), with 'You' vs 'Coach' audio comparison buttons. The AI flags specific sounds in red/yellow within the sentence text.
• Analyzed the post-assessment flow: SpeechScan identifies focus areas, locks additional focus areas behind '3 more days of practice' — a smart retention hook. The skill tree (1-6) unlocks progressively, creating a Duolingo-style progression system.
• Reviewed 70,000+ App Store reviews (4.8 stars) to identify recurring pain points, what users love, and feature gaps.

Outcome

Three things I'd change: 1. Show a before/after audio comparison after each session. Scores are helpful, but hearing your own improvement is way more motivating. The compare feature already exists — it's just buried. Surfacing it at the end of every session would make progress feel real. 2. Add a 'read and practice' mode in AI Chat. Right now it's free conversation, but people naturally avoid sounds they're bad at. If my TH scored 58%, give me sentences packed with TH sounds to read out loud — 'The weather this Thursday will be thirty degrees.' The SpeechScan data is already there, just use it to generate targeted sentences. 3. The onboarding question 'How often do you speak English?' should let you pick multiple contexts instead of one. Most people use English for work and socially, and the stakes are different in each. Multi-select would let the app personalize better — including which Role-Play scenarios to show first.

What I Learned

I went through App Store reviews to see what real users say. A few patterns came up repeatedly: What users love — the sound-level feedback and the compare/record feature. These are the core of the product and people come back for them. Common frustration — the AI sometimes keeps flagging the same sound as wrong even after the user has improved. It feels like hitting a wall, and I can see how that kills momentum. Feature gap — several users asked for the ability to choose their own words to practice, instead of only working through coach-selected ones. Overall the product is strong. 4.8 stars and 5M downloads speaks for itself. The gaps are small but the kind that affect whether someone practices daily or drops off.