Singsound Turns to AWS to Handle Workload Spikes

Millions of students in grades K–12 rely on Singsound to assess their English speaking and writing capabilities. The company’s artificial intelligence (AI) platform runs on the cloud and processes up to 160 million tests that measure the speaking and writing competencies for as many as three million students every day who attend over 300 educational institutions, primarily in China.

The Singsound platform experiences heavy workloads between the hours of 5 and 10 PM, making that time window critical for ensuring students receive fast application performance as they take speech and writing tests. If the cloud infrastructure does not deliver questions and test results quickly, Singsound risks losing customers.

As the company grew rapidly following its launch in 2016, Singsound’s cofounder and chief executive officer, Long Qin, realized the company needed to expand its cloud environment to keep up with increased student testing demands. The company initially relied on an alternate cloud provider based in China, but students encountered degraded web performance, including tests taking a long time to load.

This issue grew particularly acute when seasonal retail activity in China spiked and other customers sharing the same cloud provider generated excess traffic that negatively impacted compute resources shared by Singsound. In some cases, students couldn’t access tests at all.

“The provider suggested we purchase more compute resources to keep on standby for these busy time periods, but that would have increased our costs considerably,” says Qin. “Yes, we needed to improve the reliability of our services, but we wanted to do so cost effectively to maintain a strong cash flow.”

To resolve the performance issue, Qin turned to Amazon Web Services (AWS) for additional support. AWS had provisioned a cloud environment for a previous company that Qin worked for, so he knew he could trust AWS to deliver consistent application performance. Qin also knew AWS could provision compute resources at a reasonable cost.

Another key factor that attracted Qin was the AWS EdStart program. AWS EdStart helps startups build innovative teaching and learning solutions in the cloud while also providing qualifying companies with promotional credits for AWS Cloud services. With AWS EdStart, Singsound can access consultation services, training, and community forums with other education startups.

Since joining AWS EdStart, Qin has been actively participating in and attending community forums and education bootcamps with other members. At a recent event, Qin presented insights to other AWS EdStart Members on how Singsound leverages AI.

“The AWS EdStart program helps us connect with customers and partners,” Qin points out. “It’s a great way to network within the education sector.”

The Singsound cloud environment uses Amazon Elastic Compute Cloud (Amazon EC2) server instances with autoscaling capabilities and Elastic Load Balancing to distribute workloads across the servers. For object storage, Singsound uses Amazon Simple Storage Service (Amazon S3) while Amazon Elastic Block Store (Amazon EBS) provides block storage and Amazon S3 Glacier delivers data archiving with long-term backup services. Singsound monitors the environment with Amazon CloudWatch.

An AWS service Qin particularly appreciates is Amazon Polly, which turns text into lifelike speech using deep learning technology. The text-to-speech service synthesizes natural-sounding human speech with dozens of voices across a broad set of languages to build speech-enabled applications.

“We use Amazon Polly for our service that lets students listen to speakers to pick up the nuances of the English language,” Qin explains. “We typically hire actors, but they aren’t always available and can be expensive. AWS gives us a lower-cost alternative, and the technology comes very close to sounding like humans, with options for male and female speakers.”

According to Qin, setting up the IT infrastructure on AWS was easy. Singsound relies on two AWS data centers in China, and Qin knows Singsound can use data centers in other parts of the world when it expands its target markets. “We also value that AWS upgrades the compute resources regularly,” adds Qin. “They reduce the cost of older resources while maintaining the same costs for new resources. That gives us more powerful CPUs and additional memory without increasing our costs.”

With the AWS environment running in parallel with Singsound’s initial cloud provider’s environment, Singsound can now use web access points to measure which cloud service delivers the fastest response at any point in time. “The AWS environment consistently performs at a higher level,” Qin says. “The students of our customers do not experience testing issues.”

Directing students to the AWS infrastructure proved particularly beneficial when the shutdown struck, and students were forced to learn from home. The traffic hitting the Singsound site increased eightfold within the first couple of weeks of the shutdown and has remained high ever since. “Thanks to AWS, we have handled this spike without our costs spiraling out of control,” Qin says.

In addition to the technology and cost benefits Singsound receives from AWS, Qin emphasizes the key role of the AWS team: “We like the way AWS works with us and presents various services so we can clearly understand the return on our investment. They also communicate in a way that gives us confidence we get the best advice for helping our company succeed in our mission. We trust AWS—the services recommended always exceed our expectations.”