AI-powered language learning app Speak is in trouble.
Since its inaugural launch in South Korea in 2019. To talk has grown to greater than 10 million users, CEO and co-founder Connor Zwick told TechCrunch. Over the past five years, its user base has doubled every yr, and Speak now has customers in over 40 countries.
Investors, wanting Speak to proceed its expansion, are now pledging additional funds for the startup.
This week, the company accomplished a $20 million Series B extension led by Buckley Ventures, with participation from OpenAI Startup Fund, Khosla Ventures, Y Combinator co-founder Paul Graham and LinkedIn executive chairman Jeff Weiner. The capital injection brings Speak’s total to $84 million and doubles the startup’s valuation to half a billion dollars.
Speak, launched in 2014 by Zwick and Andrew Hsu, who met during a Thiel Fellowship, goals to show language by allowing users to learn speaking patterns and practice repetitions in specially designed lessons, relatively than memorizing vocabulary and grammar. In this respect, it is no different from Duolingo, and especially from Duolingo’s newer generative AI features. However, true to its verb, Speak emphasizes primarily verbalization.
“Our core philosophy is focused on getting users to speak as loudly as possible,” Zwick said. “Achieving fluency helps people connect, connect cultures, and create economic opportunities. It remains the most important part of language learning for people, although historically the least supported by technology.”
Speak began with English and has since launched Spanish lessons based on a speech recognition model trained on internal data. Next up is French, but Zwick didn’t say exactly when it will launch lessons in that area.
Speak makes money by charging $20 a month or $99 a yr for access to all of the app’s features, including review materials and one-time courses.
Employing 75 people in offices in San Francisco, Seoul, Tokyo and Ljubljana (the capital of Slovenia), Speak’s short- and long-term roadmap is to develop latest models that provide higher real-time feedback on tone and pronunciation, Zwick said.