The AI-powered language learning app Speak is a huge hit.
Speak has grown to more than 10 million users since entering its first market, South Korea, in 2019, CEO and co-founder Connor Zwick told TechCrunch. User numbers have doubled every year for the past five years, and Speak now has customers in more than 40 countries.
Investors, eager to see Speak continue to expand, have now committed to putting additional capital into the startup.
The company closed a $20 million Series B round this week led by Buckley Ventures, with participation from OpenAI Startup Fund, Khosla Ventures, Y Combinator co-founder Paul Graham, and LinkedIn board chairman Jeff Weiner. The capital infusion brings Speak's total funding to $84 million and doubles the startup's valuation to $500 million.
Launched in 2014 by Zwick and Andrew Hsu, who met at the Thiel Fellowship, Speak is an app designed to teach language by helping users learn speech patterns and practice them repeatedly in carefully crafted lessons, rather than by memorizing vocabulary and grammar. In this way, it's similar to Duolingo, and especially its new generative AI feature. But as its name suggests, Speak prioritizes verbalization above all else.
Image credit: Speak
“Our guiding principle is to get users to speak out loud as much as possible,” Zwick says. “Fluency allows people to build connections, connect cultures, and create economic opportunity. This is the most important part of language learning for people, but it's historically the part that's least supported by technology.”
Speak started with English, then launched Spanish lessons using a speech recognition model trained on in-house data. French is next, although Zwick declined to say when French lessons will arrive.
Speak makes money by charging $20 a month, or $99 a year, for access to all of the app's features, including review materials and one-off courses.
Speak, which has 75 employees across offices in San Francisco, Seoul, Tokyo and Ljubljana, the capital of Slovenia, has a short- to long-term roadmap that includes developing new models that provide better real-time feedback on intonation and pronunciation, Zwick said.