What is Voicely?

Language Fitness.
Not Language Learning.

Voicely is a language training platform built on Second Language Acquisition research. It measures fluency across four independent dimensions — and compounds that measurement into a training system that gets more specific to you every session.

Not a content library. Not a streak app. A fitness system — built for the same reason you'd hire a coach instead of watching YouTube workouts.

The core premise

Fluency is a physical skill. It obeys the same laws as athletic performance.

You don't become a runner by reading about running. You don't become a pianist by understanding music theory. You become fluent by training — repeatedly, measurably, under increasing pressure — until correct production becomes automatic.

This is not a metaphor. It is the literal finding of five decades of Second Language Acquisition research. DeKeyser (2007) demonstrated that language skill acquisition follows the same proceduralization pathway as every other motor skill: declarative knowledge becomes automatic through deliberate, structured practice. There is no shortcut.

Every other language app is a content library with gamification on top. Voicely is a training system with content underneath. That inversion is the entire product.

What makes it different

Four things no other language app does.

01

Measures accuracy, not attendance

Voicely only moves your rings when something improves. No XP for opening the app.

02

Compounds over time

HEXI routes tomorrow's training toward today's weakest point. Monday's session informs Tuesday's — permanently.

03

Dialect-specific, not generic

Québécois French is a different phoneme system from Parisian French. 33 dialects, each modeled separately.

04

Research-backed, not research-adjacent

Every design decision has a citation. The Island™, HEXI, retention architecture — all grounded in SLA literature.

FP Rings™

Four dimensions. Each independently scored. None hidden behind an average.

Traditional apps collapse your progress into a single score, level, or streak. That number hides where you actually stand. Voicely scores each of the four dimensions that constitute fluency separately — because your Pronunciation can be strong while your Production lags. Averaging them tells you nothing useful.

74
Ring 1

Pronunciation

Phoneme accuracy scored at the dialect level. Not accent reduction — exact deviations from your target phoneme system, per sound.

Why it matters

You cannot produce what you cannot distinguish. Flege (1995) demonstrated that incorrect phoneme categories, once formed, actively resist correction without targeted training.

Flege (1995) — Speech Learning Model

68
Ring 2

Comprehension

Real-time decoding of native speech at full speed, in your target dialect. Not translation — automatic processing.

Why it matters

Krashen's Input Hypothesis and Nation's (2001) research on comprehensible input establish comprehension as the gateway to acquisition. You must process, not just hear.

Krashen (1985) · Nation (2001)

52
Ring 3

Production

Speed and accuracy of sentence construction under pressure. The gap between knowing a word and deploying it in real conversation.

Why it matters

DeKeyser (2007) and Anderson's ACT-R model establish that declarative knowledge (knowing rules) must become procedural (automatic) through deliberate practice under time pressure.

DeKeyser (2007) · Anderson ACT-R (1983)

81
Ring 4

Retention

How well structures hold over time. HEXI models your personal forgetting curve and resurfaces items before they decay.

Why it matters

Spaced repetition research (Ebbinghaus, 1885 — still the most replicated finding in memory science) shows retrieval timing is the dominant variable in long-term retention.

Ebbinghaus (1885) · Cepeda et al. (2006)

Full breakdown of FP Rings™ →

HEXI™

The compound intelligence layer.

HEXI — Holistic Experience Intelligence — is the AI system that reads your four rings after every session and determines what to train next. It is not a recommendation engine. It is a compound trainer.

HEXI connects your pronunciation failures to your comprehension ceiling. It routes your vocabulary gaps into your production drills. It times your retention sessions to your personal forgetting curve — not a population average. Every session, it has a more complete model of you.

This is what compounding means in practice: Monday's session is not a repeat of last Monday's. It is a direct response to what HEXI measured on Friday, Tuesday, and four weeks ago.

How HEXI works in depth →

Ring gap analysis

After every session, HEXI reads the delta between your four rings and identifies which dimension is your current ceiling.

Cross-modal routing

Vocabulary gaps surface in listening. Comprehension failures route to production drills. Phoneme errors trigger Island™ reps. Every gap has a targeted response.

Forgetting curve prediction

HEXI predicts when each structure will decay for you specifically — not a population average — and resurfaces it at the optimal moment.

Session sequencing

Based on Anderson's ACT-R model: declarative knowledge first, procedural practice second, then integration across modes. The order is not arbitrary.

HEXI doesn't ask how you feel. It reads your rings — and already knows.

Voice Clone

Train against the version of you that's already fluent.

Record 30 seconds. Voicely builds a model of your voice speaking your target language at native fluency — your intonation, your timbre, your rhythm. That model becomes your pronunciation target for every session.

You are not imitating a native speaker. You are training toward a version of yourself. That distinction changes the neural pathway: your brain processes it as a gap to close, not a foreign identity to adopt.

No other app puts your own voice at the center of the pronunciation model. This is what makes the fluent-self concept unique to Voicely.

01

Record 30 seconds

Any content. Your voice, your natural cadence.

02

Voicely builds your model

ElevenLabs voice cloning tuned to your target dialect.

03

Your fluent self speaks

Every session, you hear yourself pronouncing correctly.

04

HEXI routes the gap

Weakest phonemes in Ring 1 get your model first.

Learn more about Voice Clone →

VOICECAST™

Your personalized fluency fingerprint.

VOICECAST is a two-minute spoken assessment that returns your four FP ring scores — not a level, not a grade, not a comparison to other users. Four numbers that tell you exactly where you stand and which ring is your ceiling.

Duolingo says B2. Babbel says Advanced. VOICECAST says: Ring 1 at 74, Ring 2 at 68, Ring 3 at 52, Ring 4 at 81 — and your Ring 3 ceiling is Production speed under conversational pressure. Here is the next three weeks of training.

Duration

2 minutes

Format

Voice recording

Output

4 ring scores

Next step

HEXI training plan

Get your VOICECAST™ score →

The research

Every design decision has a citation.
Not inspiration. A study.

Voicely is not research-adjacent. The SLA literature is the specification document. These are the six foundational findings the system is built on.

DeKeyser (2007)

Practice in a Second Language

Declarative knowledge becomes procedural through targeted practice under increasing time pressure. This is why the Island™ exists — isolated phoneme and structure drills before contextual use.

Anderson, ACT-R (1983–1993)

The Architecture of Cognition

All skill acquisition follows a three-stage sequence: declarative encoding, proceduralization through practice, and automatic production. HEXI's session sequencing maps directly onto these stages.

Flege (1995)

Speech Learning Model

L2 phoneme categories that deviate from L1 categories must be explicitly targeted. General listening is insufficient. Voicely's Ring 1 system operationalizes this finding at the individual phoneme level.

Nation (2001)

Learning Vocabulary in Another Language

Vocabulary retention requires spaced encounters in varied contexts. Single-exposure learning is almost always temporary. Voicely's HEXI retention engine is built on this principle.

Skehan (1998)

A Cognitive Approach to Language Learning

Individual differences in language aptitude, working memory, and learning style produce radically different acquisition trajectories. Generic lesson sequences are guaranteed to fail most learners.

CEFR Companion Volume (2020)

Council of Europe

Mediation and interaction competencies — performing under communicative pressure — are distinct skills from receptive competencies. Most apps only train the latter. Voicely trains all four.

Read the white papers →
M

Where it came from

Mervine Gowry

Founder, Voicely Language

Eight years teaching French online — 60K TikTok followers, a YouTube channel, language courses — and the persistent observation that even motivated, consistent learners plateau. Not because they lacked effort. Because every tool they were using was optimising for engagement metrics, not acquisition outcomes.

Voicely is the system I wished existed. Built on the SLA literature, not around it. Every feature has a mechanism behind it. Every design decision has a study. The research is not a justification for choices already made — it is what determined the choices.

Full founder story →

Ready to stop learning
and start training?

Two minutes. Your four FP ring scores. A HEXI training plan built specifically around your weakest dimension.