How scoring works

Voice Score is a composite of measured acoustic and pacing features compared with reference ranges for clear, confident speech, as reported in listener studies of English-speaking audiences. It is not a clinical, diagnostic or medical measure, does not measure ability, and does not measure attractiveness as perceived by any specific person.

What is measured

Your iPhone records about 20 seconds of voice — a held vowel and a short read passage — and analyses it on your iPhone: median pitch and pitch movement, voice steadiness (cepstral peak prominence and harmonics-to-noise ratio), loudness steadiness, speaking rate against the passage's known syllable count, and pausing. The recording itself never leaves your phone; only these numbers do.

How the number is built

Each measurement is compared with reference ranges for clear, confident speech and combined into subscores and one composite using fixed, published weights. The composite never depends on the transcript — it is identical in every language mode and never changes after your reveal. Every score carries an uncertainty band: single recordings vary, so treat week-over-week movement inside the band as noise, not change.

Calibration

The reference ranges and weights come from published speech research and are calibrated against recorded speech: our own reference recordings, and openly licensed speech datasets — from professional narration to clinician-rated voice samples — as each calibration lands. When a calibration ships, the app notes the rubric version on your history so numbers are always compared like-for-like.

Coaching

Coaching tips are generated from your measurements and transcript (never audio) by Anthropic PBC's models — or Google LLC's if Anthropic is unavailable — only when you turn AI coaching on. AI-generated coaching. Not medical or professional advice.

This is a communication-skills tool, not a medical or psychological assessment. You must be 13 or older.