Isthmus. Nuclear. Anemone. Tricky English pronunciations are a challenge for many immigrants to the U.S. and native-born speakers alike. ELSA — which stands for English language speech assistant — aims to help with that.
The three-year-old Silicon Valley startup offers an English-pronunciation app, dubbed ELSA Speak, that’s geared toward American English and available for Android and iOS devices.
Vu Van, ELSA’s co-founder and CEO, said English pronunciations plague many seeking careers. A Vietnamese immigrant, Van picked up English early and later went to Stanford University to earn an MBA in 2011, but still struggled with certain words, leading her to start ELSA. 
ELSA’s app is designed to be a personalized coach for practicing English, particularly for non-native speakers. It offers bite-sized lessons intended to improve pronunciation with 10 minutes a day of practice.
While most language apps emphasize grammar, “we are very focused on pinpointing your pronunciation errors,” Van said. “It’s supposed to help people with accents.”
She said that having an accent can crush one’s confidence and that working on pronunciation is difficult without the aid of an expensive tutor.
That’s where ELSA comes in. The app uses AI and speech-recognition technology to help people practice English for professional and everyday situations.
Practice Makes Perfect
The coaching app, which enables people to set daily practice reminders, has a slick interface that makes learning easy and fun. The app coach shows a sentence and prompts you to tap the microphone icon and say it. It gives positive feedback in bold, writing EXCELLENT in big green lettering for good pronunciations.
Perhaps even more valuable, it counters mispronunciations with helpful tips to get it right. For example, the language coach offers a number of pointers that help users understand where to place their tongue in their mouth and how to hold their lips when saying particular words.
ELSA is geared to help non-native speakers focus on sentences commonly used in a new job or at a conference, among other professional settings.
The app first takes people through a five-minute assessment test to identify challenges. It offers more than 600 two-minute English lessons and more than 3,000 words for people to practice.
Users’ conversational English lessons are recorded and scored in the app to help gauge their pronunciation level on specific words.
ELSA’s Coaching Evolution
To train its pronounciation model, the company fed thousands of hours of spoken English into a recurrent neural network. It’s now fine-tuning its algorithm and is constantly training with data from users of its app, said Van.
“The more NVIDIA GPUs we have, the more experiments we can run on the model,” she said.
Launched in 2016, ELSA’s apps have been downloaded more than 2 million times. The service is free for the first week, and then requires a subscription to continue beyond limited access. Subscriptions cost $3.99 for a month, $8.99 for three months or $29.99 for a year.
ELSA is a member of the NVIDIA Inception program, a virtual accelerator that offers hardware grants, marketing support and training with deep learning experts.
The startup recently scooped up $3.2 million in venture funding. The founders are seeking additional AI talent to help further build out the service.