OpenAI-backed Speak raises $78M at $1B valuation to help users learn languages by talking out loud

Languages are typically taught by first exposing you to reading and writing, but native speakers always start learning languages by hearing and talking. Speak has built a platform to teach languages by focusing on how native speakers learn: Using AI, the startup generates audio conversations and listens to users’ responses to improve their grasp of a language.

Now, Speak is announcing a milestone fundraise that further bolsters its progress: a Series C of $78 million that catapults its valuation to $1 billion. 

The funding round is being led by Accel, with previous backers OpenAI (via its Startup Fund), Khosla Ventures and Y Combinator also participating. 

The fundraise is a major jump forward for the startup. Speak confirmed a Series B extension of $20 million at a $500 million valuation only six months ago. 

Some of this investment activity may well be down to the huge enthusiasm around generative AI and one of Speak’s key investors. OpenAI, as you might have guessed, is not just a financial backer here. Speak is using the company’s technology to power its platform, and the ChatGPT maker’s also an early partner for the latest in its speech technology. By association, Speak is proving one of the commercial opportunities for GenAI. 

“Our 2022 investment in Speak was driven by a shared vision of revolutionizing language learning with AI,” said Ian Hathaway, a partner at OpenAI’s Startup Fund, in a statement. “Weʼre thrilled to see their world class AI talent and unique product vision create transformative learning experiences for a rapidly growing userbase worldwide.”

One of the primary aims for the funding will be to expand the number of target languages users can learn with Speak — and thus its customer base — starting with Spanish and French. 

So far, eight-year-old Speak has been focusing on people studying just one language — English, the world’s most popular language for learning. Speak provides learning and review materials designed around courses; the idea here is that it will complement what users may have studied elsewhere. Currently the company lists eight originating languages for learning English, based on the most popular language groups among English learners to date.

“For the one and a half billion people out there trying to learn English, the vast majority of them have spent 15+ years learning intensely. They know vocabulary and grammar better than any of us. But the issue is that they have no ability to speak it,” said CEO Connor Zwick, who co-founded the company with Andrew Hsu (CTO). “For us, our value proposition up until now has really been, let’s teach people how to communicate in the language.”

To be clear, that 1.5 billion figure is Speak’s total addressable market, not its user number. It doesn’t actually disclose how many active users it has. As a guide, Hsu said Speak’s app has been downloaded more than 10 million times, and average usage is around 10-20 minutes/day, paying $20 per month, or $99 per year, typically a fraction of the price of hiring a human tutor to work on improving conversation.

Speak for Business, an enterprise tier, has over 200 customers, the company said. 

Zwick describes Speak as part learning method and part tech platform that works in a three-step process. 

First, you are thrown into listening and talking — an interesting approach, considering that Hsu and Zwick met and started working together after going through a cohort as Thiel Fellows, where you are, in theory, thrown into building an enterprise rather than going through years of learning first.

“We’re not going to explain every single grammatical rule to you,” Zwick said of the first step of its program. 

Second, you are then asked to apply that new term or phrase over and over — “basically drills where you just practice saying it out loud with various other pieces of languages so that it becomes automatic, with no translation.” 

Third, Speak then presents the phrase in “a real-world context using AI… That’s how you really anchor it,” Zwick said. 

Ironically, although the aim is to get its learners speaking a new language with humans, there are no humans in the loop. This is all crafted using speech recognition, natural language processing, generative AI and more to tailor the learning to the learner. 

Does it work, though? For now, Speak doesn’t have any integrations with any standardized language learning qualifications, if you believe in that particular metric. That is a route others in the online language learning space have taken, with Duolingo, for example, providing an English test that international students can use to prove their English competency for thousands of English-language universities.

“We are explicitly trying not to be a test prep solution, frankly, because, unfortunately, all tests so far are imperfect,” Zwick said. “What ends up happening is that people end up gaming tests. They try to become good at taking that test. They’re not trying to become good at actually being able to communicate and use the language in the real world. The only way to really have a proper assessment is to have an expert have a conversation with you. There are some tests like that. But how do we scale that to everyone?”

Hsu hinted that this could be part of the company’s plan in the long term, however. “This system we’re building right now to quantify fluency and efficacy, I think [it] will be really useful for launching something like a true, actually accurate English fluency score or test,” he said. 

Gamification is also not an area that Speak has explored to date, sitting out one of the bigger trends in online learning in the last several years (for now, at least). Duolingo and companies like Kahoot have leaned into it though, turning the art of learning into something of a game. New services like Eleven Labs’ multilingual AI agents launched in November will potentially open the field to more language-learning services also pursuing ways to encourage speaking that will take on features like gamification.

Some of the funding might see more consumer-driven models like this introduced as well. 

Now with more staff, “there’s room to bring more of those behavioral mechanisms into the app to result in positive change for the users,” Zwick said. But that won’t be at the expense of actually learning. “When there’s a tug of war between gamification and engagement and efficacy, we will pick efficacy 100% of the time,” he said.

Ben Quazzo, a partner at Accel who led the investment round, will be joining Speak’s board of directors. “Speak has emerged as a standout player in consumer AI, demonstrating exceptional growth and market potential,” he said in a statement.