UMSI student uses AI to improve online learning for non-native speakers

When Mnatsakan Sharafyan began his Master of Applied Data Science at UMSI, he found himself pushing buttons during online lectures. Specifically, Sharafyan kept clicking to adjust the playback speed or rewind sections of the lecture he hadn’t comprehended. As a non-native English speaker — Armenian is his first language — Sharafyan was grateful for the opportunity to watch recorded lectures at his own pace. But he quickly realized a downside: an hour-long lecture could take him three hours to complete. 

“I decided to solve this problem by using knowledge gained from my MADS studies,” said Sharafyan. 

He developed a Chrome extension, Saima, that uses AI to auto-adjust video speed. In collaboration with a neuroscientist, Sharafyan has created the first technology of its kind. Saima has the potential to improve students’ focus, comprehension and memory by maintaining the optimal rate of information delivery. 

During an online lecture, a professor might take a long pause for emphasis or speak more quickly when summarizing information. But the playback speed feature on YouTube and other video platforms doesn’t account for variations in speech rate, instead dividing or multiplying the speed by a single factor over the entire video. This is why Sharafyan had found himself adjusting the playback speed constantly during a single lecture. 

By contrast, Saima adaptively controls a video’s speed so that the rate of words per minute remains constant. Users have the option to select their desired WPM using a sliding control, with no further adjustments needed. Saima can recognize silences during a lecture and automatically increase the playback speed by 3x. For non-native English speakers, a feature called “voice boost” applies real-time audio processing techniques to improve vocal clarity and maintain consistent volume, increasing comprehension without the need to slow the video down.

While Sharafyan developed Saima with non-native English speakers in mind — an estimated 1.1 billion people worldwide — he has realized the technology holds broader potential for online learners. For example, students with ADHD might benefit from the “focus boost” feature, which increases video speed imperceptibly to hold a user’s attention. 

Sharafyan said that all of his MADS courses — from data science ethics to deep learning — “played an important role” in developing Saima, which sits at the intersection of data science and cognitive science. 

One hundred beta users at several universities, including U-M and UC Berkeley, provided positive feedback on the beta version starting in the fall of 2022. After making improvements, Sharafyan and co-founder Shushan Nersisyan launched Saima on the Chrome Store in May, with a free version that allows up to 250 minutes of video speed control per month and options for paid subscriptions. 

Over 6,000 people have downloaded Saima so far. The extension is currently compatible with YouTube, Coursera, Udemy and many other online learning platforms. Sharafyan sees the potential to expand this technology to the entertainment industry, “using computer vision models to analyze the picture of the video,” so that viewers who typically turn on subtitles to comprehend dialogue could instead watch movies at their own most comfortable speed.

“Pursuing a MADS degree at UMSI was one of my best decisions,” said Sharafyan, who plans to graduate in the spring of 2024 both as a founder and user of Saima. He strives to iterate in the same way he now listens to lectures: “smart and fast.”  

Learn more about the Master of Applied Data Science program.