Innovation of the Week: Precision pronunciation powered by AI | Crain's San Francisco

Innovation of the Week: Precision pronunciation powered by AI

More than 7,000 unique languages are spoken by the Earth’s 7.5 billion people. Mandarin is spoken more than any other language, but 20 percent of the global population — 1.5 billion people — speaks fluent or partial English. And according to language reference publication, English takes the top spot as the language most commonly studied the world over, making its education market huge.

Popular tools such as Duolingo, which is free, and Rosetta Stone, which is not, offer mainly vocabulary and grammar lessons for those looking to learn languages for the first time. But a new app called ELSA separates itself from the pack by taking a different approach.

Its creators bill the app, which uses speech recognition technology, as a “pronunciation coach” that helps language learners improve their spoken English.

ELSA is for people who already have the basics of English down but are looking to improve their pronunciation, inflection and accent. Upon signing up for the app, users take a speaking assessment so the program can determine what areas they need to work on.

The app’s creator, Vu Van, was inspired by her own experience with the English language. She was born and raised in Vietnam and moved to Denmark to work before making the leap Stateside. She settled in the Bay Area and earned an MBA and a master’s degree in education from Stanford University.

Although English isn’t Van’s native language, she says it has always been a major part of her life, both personally and professionally. When she first arrived in the United States, her proficiency in English was working but not quite fluent.

While attending Stanford, Van realized this gave her a major disadvantage in the classroom. Her thick accent made it difficult for professors and peers to understand her ideas and contributions. She found herself passed over time and again for students whose primary language is English.

“When you speak with a strong accent, people unconsciously trust you 30 percent less,” Van says. “And that bugs me a lot.”

First, she looked into hiring a speech therapist to assist her, but it was too expensive. YouTube tutorials were sparse. Luckily, Van had a friend who taught English and could help her out one-on-one. But Van saw a gap in the education market for so many others with similar speaking problems but without the same access to resources.

It was then that she had the idea for ELSA.

“More than 90 percent of [English-language learners] actually struggle a lot with their speaking,” Van says. “That’s why we built ELSA using voice-recognition technology: to be that personal speech therapist, listening to every single part of your speech and identifying what errors you made, what vowels and consonants you don’t pronounce correctly.”

With her background in education and business, Van teamed up with CTO Xavier Anguera and spent six months developing the app’s technology. Elsa has racked up more than 2 million users since its 2016 launch.

To ensure they’re offering the best possible education, Van and Anguera do a lot of data analysis with users. On average, a user spends about 10 minutes per day on the ELSA app for three to six active months. This adds up to 30 percent overall improvement, according to Van.

Right now, the main focus of ELSA is as a direct-to-consumer product, but it has already been picked up by educators. A pilot program in Vietnam has eight schools using ELSA as a supplemental homework tool, with the hope of expansion to more mainstream use.

Tony Ngo, chairman and co-CEO of Everest Education, counts his company and ELSA as part of a small circle of innovators in the Vietnamese ecosystem. He reached out to Van to team up and bring ELSA to the Fulbright Everest Launchpad program during the summer of 2017. He was drawn to the app’s unique approach and classroom adaptability.

“With the most data for native Vietnamese learning English, ELSA's AI database has the best available pool to feed their algorithms. This makes the pronunciation feedback more accurate for students and frees up teacher time [for] other tasks,” Ngo says.

Binh Tran is a venture partner at 500 Startups, one of the first investors in ELSA. He specializes in Silicon Valley-level tech companies with a focus on Vietnam.

Tran says the app helps solve “non-Silicon Valley problems — that is, problems most Silicon Valley entrepreneurs know nothing about. In this case, it’s a problem that most Americans don’t even know about.”

The goal for ELSA is to expand the app to more languages, with the tentative launch of Spanish to come in late 2018 and Mandarin to follow.

For now, the ELSA team is working on advancing their conversational AI, with simulated scenarios such as ordering coffee.

“We really want to build confidence in [users’] speaking,” Van says, “because the more confident they are when they speak, knowing that people understand them, the higher the chance they will keep correcting their mistakes.”

February 13, 2018 - 1:36am