PROBLEM: 66 phonics sounds shipping in the L2R Android app. How do I verify each one is actually pronounced correctly before it ships to a kid?
WHY IT MATTERS: A phonics app teaches kids the WRONG sound if any audio clip drifts. Manual ear-check at scale is the only ground truth — and that needs a tool a non-engineer (me, on a phone, between meetings) can drive.
HOW TO USE THIS DASHBOARD: Click any phoneme tile to hear it. Compare side-by-side against the older audio (qa_audio = pre-fix; res_raw = currently shipped in app). Approve / reject feeds back into the regeneration pipeline.
STACK: GCP TTS (Neural2-F voice for new audio), librosa for QA features (silence_pct, duration_ms), this HTML+JS dashboard for ear-check verification
Phoneme Audio QA Dashboard
Sound Bloom: Kids Phonics — Generated 2026-04-02 via GCP TTS Neural2-F
Click each phoneme to play. Approve or reject. Compare against current bad audio.
Single Letters (a-z)
Digraphs & Blends
Side-by-Side Comparison (New vs Current)
Compare the new GCP Neural2 audio against the current bad qa_audio files for selected letters.