Unique Research · Company · December 11, 2024
Language teaching usually starts with reading and writing; native speakers start with listening and speaking. Speak builds for the native path: AI-generated audio conversations that listen to user replies and lift fluency.
Milestone: $78M Series C at a $1B valuation, led by Accel, with early backers OpenAI Startup Fund, Khosla Ventures, and Y Combinator. Only six months after a $20M Series B extension at $500M — a sharp step-up. OpenAI is more than capital: Speak runs on OpenAI tech and was an early partner on its latest voice stack — a live GenAI commercial case.
OpenAI Startup Fund partner Ian Hathaway: the 2022 investment rested on a shared vision to reinvent language learning with AI; world-class AI talent and a distinct product vision are delivering transformative experiences for a fast-growing global base.
Use of proceeds: expand target languages beyond English — Spanish and French first. For eight years Speak focused on English learners (the world’s most-studied language), with course-based materials meant to complement other study. Eight source languages for learning English, keyed to the largest learner pools so far.
“1.5 billion people are learning English; most have studied hard for 15+ years — better vocab and grammar than any of us — but they can’t communicate,” says CEO Connor Zwick (co-founder with CTO Andrew Hsu). “Our value prop so far: teach people to communicate in English.” That 1.5B is TAM, not users. Hsu: 10M+ downloads; ~10–20 minutes average daily use; $20/mo or $99/yr — a fraction of human conversation tutors. Speak for Business: 200+ enterprise customers.
Three steps: (1) drop into listen/speak practice without explaining every grammar rule; (2) drill the new word/phrase aloud until automatic, without translation; (3) AI places the phrase in real-world context — “that’s how you really own it,” Zwick says. No humans in the loop — speech recognition, NLP, GenAI customize content. Ironic goal: talk to humans; process: all machine.
Does it work? Speak is not yet tied to standardized language credentials (unlike Duolingo’s English test for universities). “We’re explicit: we don’t want to be a test-prep solution — tests so far are imperfect; people get good at tests, not at real-world use. True assessment is expert conversation — how do we scale that to everyone?” Zwick said. Hsu: systems to quantify fluency could later support real English-fluency scores.
Gamification is largely untouched so far (Duolingo, Kahoot own that trend); ElevenLabs’ multilingual AI assistant and peers may push more speak-first, game-like products. Some capital may fund consumer engagement mechanics — “when gamification fights effectiveness, we pick effectiveness 100%,” Zwick said. Accel partner Ben Quazzo joins the board; Accel calls Speak a standout consumer-AI name with exceptional growth and market potential.
— END —