Loora, a generative AI app that uses an audio interface to help users learn English, raises $9.25M

The jury is still out on whether AI adoption really will spell the end of humanity as we know it, but in the meantime, we’re continue to see a boom of AI startups and AI dealmaking. In the latest development, a generative AI-based language learning startup called Loora is coming out of stealth with $9.25 million in seed funding to improve people’s conversational English skills, no matter the topic, by way of on-the fly, voice-based interactions with the eponymous Loora iOS assistant.

Unlike many of the majority of generative AI applications right now that are text-based — you write out prompts for ChatGPT, Midjourney, and the rest — Loora is completely focused on audio. That is to say, you speak to it to learn, as you might do with a human tutor, and Loora speaks back to you. The startup claims that its AI tutor can understand and help improve anyone’s bad English, and bad English accent, across any topic under the sun.

The funding is coming mainly from three early-stage investment specialists, Emerge leading the round, with Two Lanterns Venture Partners and Kaedan Capital, and various angels participating.

Tel Aviv-based Loora, started in 2020, may be coming out of stealth today, but Roy Mor, the CEO who co-founded the company with Yonti Levin, noted that it’s been quietly operating for a while now and has already signed up “thousands” of paying customers — consumers and prosumers who want or (for work) need to improve their English-speaking skills and are using Loora’s app to do so.

And in turn, Loora is using that interactivity to continue expanding its knowledge base.

Mor, a machine learning alum from Mobileye, said that he and Levin (Loora’s CTO) decided to focus on language tuition, and English specifically, for a couple of reasons.

The first is that English is, by far, the most common language being learned right now, with more than 1 billion learners globally. (And the plan is to stay around English for now.)

The second is that while there are dozens (if not more) popular online language-learning platforms out in the world already, they found a gap when it came to mastering conversational English in a one-to-one learning environment.

At one end are language learning apps. Platforms like Duolingo, Memrise and Babel, Mor said, “do a great job but they mostly cater to a casual audience.” A typical user might just as easily be on TikTok as on one of these apps, and it might give you basic skills, but not much more. “Being able to order a coffee is great,” said Mor, “but if you want to work in English, it’s not relevant enough.” And they are limited around what they can do with the voice-based interactions that they do have.

At the other end are human tutors, either in person or online through video meetings, who can hone in on specific topics that you might want or need to cover, listen to you and shape their teaching based on your particular needs. “But for many, those actual tutors are prohibitively expensive,” he explained.

The solution, inevitably, has come by way of advances in tech. Leaning on the new wave of generative AI — popularized by services like ChatGPT form OpenAI and the Midjourney illustrator that creates images based on verbal prompts — Loora has built an English-language tutor that responds to conversation prompts, comes up with conversation prompts when you don’t have one to mind, and understands and responds to what you say to it in order to give you guidance on how to improve your spoken English.

Mor said that initially, in 2020, Loora built its own large language models. But now that these have become widely available it’s tapping a variety of them for its service — some are suited better to different tasks, so different LLMs are used, for example, for openers (when someone starts using the app), for first conversations, subsequent conversations, different topics, and a different one for the app’s “memory” of each user. Loora uses these third-party LLMs, he said, “but we train on our own data.”

To be clear, Loora is not ingesting audio per se — that is still a very nascent area, with companies like Meta launching Voicebox, its first effort in that space, only earlier this month; and others in academia just starting to look at how and where voice-based-input and interfaces might work. But it is using a mix in which the audio is an essential component, given that the aim is to improve spoken English.

“We use a speech-to-text model since our conversational and grammar feedback models rely on textual input,” Mor said, but along with that, “We also use the speech itself in order to assess and give feedback on pronunciation, prosody, and fluency, as well as understand the sentiment.”

That also means there is an inherent limitation to the app as it exists: Loora is focused on speaking, not reading and writing, and it seems that you have to have at least some knowledge of English — rather than none at all — to get started. There will be some reading introduced, Mor said, with an upcoming module that will provide texts to read as the anchor of a subsequent discussion. But writing is not really on the table right now.

This means that although Loora is certainly setting out to do something different from what other language learning apps on the market provide now, if you want to master a language, as with the others, you will still likely need to use it in tandem with other learning experiences.

Nevertheless, Loora’s arrival is seems like a notable advance in what’s out in the market today, and longer term likely to be just one of the many efforts to advance audio interfaces for generative AI services.

“We have been partnering with trailblazing AI companies since 2016, so Loora was a natural fit in our portfolio. In 2020, co-founders Roy Mor and Yonti Levin already recognized the potential of generative AI for language learning and set out on a mission to harness it for the benefit of society, far before the introduction of ChatGPT,” said Liad Rubin, general partner at Emerge, in a statement. “General-purpose generative AI models are effective in the execution of generic tasks, but there is a massive advantage in building and training an AI specifically for a domain. This is why Loora envisioned and built their AI tutor with the sole purpose of teaching English, creating an unparalleled, next-gen solution for virtual English education. We’re excited to be part of this journey.”