Chivox AI builds speech assessment and pronunciation-scoring technology for language-learning products, AI tutors and voice agents. Two decades of R&D — from Cambridge origins to a NetDragon company — are exposed through APIs and MCP tools that product teams can integrate without rebuilding the speech stack.
Chivox (驰声) is a speech-technology company specializing in intelligent pronunciation assessment. The team began in 2007 from Cambridge University speech work, then built proprietary scoring engines on statistical modeling, NLP and large-scale learner audio.
Since joining NetDragon (HKEX: 0777) in 2015, that stack has powered language-learning apps, publishers, smart hardware and exam workflows. Today the same engines ship as API, SDK and MCP tools — covering 260+ English and Mandarin item types, with production reliability designed for real classroom and product traffic.
International work includes Thailand Ministry of Education smart-classroom collaboration and Japan ESAT-J exam prep products — proof that assessment detail can travel across curricula without rewriting the integration model.
“We hope that with the development of artificial intelligence, language teaching and learning will become more and more efficient.”
The founding team started from Cambridge University speech research, with early support from UK R&D and university funds.
Speech scoring moved into formal English listening-speaking exams — the bar that still shapes how we treat reliability and fairness.
Chivox joined NetDragon (HKEX: 0777), expanding distribution into global education products and platforms.
Opened a Beijing education-exam research institute; supported Thailand MoE smart-classroom work and Japan ESAT-J prep products.
Two decades of engines exposed as SDK, API and MCP tools for tutors, publishers, hardware and voice agents worldwide.
Chivox turns deep speech research into production-ready pronunciation scoring: detailed enough for diagnosis, structured enough for product logic, and consistent enough to operate across languages and markets.
A score is useful. Inspectable evidence is what makes the next product decision possible.
Pronunciation, fluency, tone and audio-quality signals are developed as assessment capabilities rather than prompt-level guesses.
Stable response fields let teams combine deterministic product rules with model-generated explanations and learner feedback.
The same foundation supports different learner groups, connectivity conditions, curricula and business models without flattening language detail.
Office, lab and team photos from our Chinese careers page. See how we work, what we offer, and how to apply.
View culture & careers
The same assessment engine runs behind global learning platforms, classroom hardware and publisher content — and behind teams shipping offline apps, school platforms and online courses with the SDK and MCP today.
An offline English-learning app for learners in connectivity-constrained regions, sold as a one-time purchase with lifetime access and focused on IELTS, TOEFL and Cambridge English listening and speaking practice.
Visit websiteA school-focused technology company expanding from attendance and security systems into a Chinese and English learning platform.
Visit websiteAn online Chinese-learning provider building a structured Mandarin and HSK course experience for local learners.
Visit websiteTell us what you’re building and whether you need SDK/API, MCP, or function calling. We’ll reply within one business day with pilot credits, pricing, or a deployment plan.
