Product · Mistral AI Voxtral powered
Talk to an AI English coach. Get scored in real time. Progress from A1 to C2 with a CEFR-calibrated voice partner — built on Mistral AI Voxtral for listening and speaking, and a Mistral chat model for coaching.
Adult-oriented professional learning product. Not designed, marketed, or distributed for children under 13.
What it is
VoiceCraft© listens, transcribes your speech with Voxtral, sends it to a Mistral chat model that plays the role of a CEFR-level-aware language coach, streams the response back, and reads it aloud with Voxtral TTS — all in real time.
After each turn it scores grammar, vocabulary, pronunciation, and coherence, and updates an ELO-style rating that tracks your progress through the CEFR bands.
Who it is for
Capabilities
Voxtral STT
Streaming audio is transcribed through Mistral Voxtral transcription models. Auto end-of-utterance detection via browser RMS energy, plus a manual stop control for noisy environments.
Mistral chat
Six distinct system prompts, one per CEFR level. Tone, vocabulary ceiling, and correction style adapt from patient beginner guidance (A1) to challenging C2 interlocution.
Parallel feedback
Each turn is analysed in parallel with the coaching reply. You get per-dimension scores and targeted corrections — not a generic thumbs-up.
Voxtral TTS
Streaming coaching replies are spoken back with Voxtral TTS. A neutral articulate English preset ships by default; zero-shot voice cloning is available for enterprise deployments.
ELO proficiency
A 0–100 ELO-style rating moves after every interaction. Higher K-factors at A1/A2 accelerate early wins; lower K-factors at C1/C2 make mastery earned, not given.
French L1 tuning
Coaching prompts target false friends, article interference, preposition choice, and question-inversion calques specific to French L1 speakers acquiring English.
CEFR coaching personas
Each CEFR band has its own system prompt, tone, vocabulary ceiling, and correction policy. The coach you speak to at A2 is not the same coach you speak to at C1 — and the jump between them is earned, not declared.
Breakthrough
Patient · simple vocabulary
Waystage
Concrete topics · short exchanges
Threshold
Everyday work · clear prompts
Vantage
Business fluency · nuanced feedback
Effective
Professional register · strict scoring
Mastery
Sophisticated interlocutor · challenges
Grammar
40%
Vocabulary
25%
Pronunciation
20%
Coherence
15%
How it works
01
You speak
Audio streams over WebSocket to a FastAPI backend in small chunks.
02
Voxtral transcribes
Mistral Voxtral returns a text transcript of your utterance.
03
Coach responds
A Mistral chat model streams a CEFR-calibrated spoken reply.
04
Feedback scored
In parallel, a second Mistral call grades grammar, vocabulary, and coherence with corrections.
05
Voxtral speaks
The reply is synthesised back to you via Voxtral TTS and your ELO rating updates.
Conversation topics
Each topic ships with target vocabulary, grammar focus points, and varied starter prompts, so every session begins with purpose rather than small talk.
Travel
Business
Culture
Technology
Health
Debate
Education
Personal
Age policy
VoiceCraft© is designed for adult and professional learners. It is not intended, is not marketed, and is not distributed for audiences under 13. Content, tone, topic range, and feedback calibration assume an adult professional context. Institutional deployments to under-18 learners require an education licence under BrightNTech.AI terms with appropriate supervision and safeguarding.
Ready to try VoiceCraft©?
Request access for yourself, or speak to BrightNTech.AI about licensing VoiceCraft© for your team, university, or language programme.
FAQ
VoiceCraft© is a real-time AI voice coaching product that listens to you speak, scores your English across grammar, vocabulary, pronunciation, and coherence, and holds a spoken conversation back in a CEFR-calibrated coaching persona from A1 to C2.
VoiceCraft© runs on the Mistral AI stack: Voxtral for audio transcription, a Mistral chat model for streaming coaching dialog and parallel feedback analysis, and Voxtral text-to-speech for spoken responses.
VoiceCraft© is built for adult and professional learners — in particular French L1 speakers acquiring business and conversational English. It is explicitly not a children’s product, does not target audiences under 13, and is not marketed through children-facing channels.
Every turn is scored on grammar, vocabulary, pronunciation, and coherence. An ELO-style rating updates after each interaction, with higher K-factors at beginner levels for faster encouraging movement and lower K-factors at advanced levels so progression to C1 and C2 requires sustained performance.
VoiceCraft© ships with 20 curated conversation topics across 8 domains: travel, business, culture, technology, health, debate, education, and personal narrative. Each topic carries target vocabulary, grammar focus points, and varied starter prompts.
VoiceCraft© is a proprietary BrightNTech.AI product commercialised by 7Milestone SASU (Paris, France). Access is available through subscription on iOS and web, with enterprise and education licences available on request via BrightNTech.AI.