How this AI startup is overcoming the challenges of cultural accents

New technology for help overcome challenges with understanding individual accents could have huge implications for marketers looking to improve customer service experiences

Andrés Pérez Soderi has a joke he likes to tell when asked to describe how his company came together.

“A Russian, a Chinese and a Venezuelan walked into a bar, and they end up making software so they can understand each other,” Soderi says.

Venezuelan-born Soderi is one of the three founders of US-based artificial intelligence startup, Sanas. Together with Chinese-born co-founders, Shawn Zhang, and Russian-born, Maxim Serebryakov, and their team, he has been working to unpick the nuances of accents to make communication easier for people from different backgrounds.

That could have huge implications for any marketer who wants to improve customer experience through voice channels by helping both customers and staff be better understood.

Soderi tells CMO the true inspiration for Sanas came from the experience of a mutual friend working in a contact centre in Guatemala, whose interactions with customers were constantly hampered by their inability to understand his accent.

“This guy was one of the smartest people we knew,” Soderi says. “The fact he wasn’t able to hold down a job in a customer service centre simply because of his accent, we didn’t think was fair. And accents aren’t something you can very easily change.”

Saanas’ solution uses custom algorithms to interpret incoming speech and provide a ‘translated’ output in near-real time. This audio output can be fed directly to the contact centre agent to improve their understanding of the incoming caller. The voice of the agent can also be translated into an accent that might be more easily understood by the caller.

“We don’t neutralise accents, we fully convert them or translate them,” Soderi says. “That means we can do German to Australian, or Spanish to Hindi – you can essentially do pretty much anything.

“If you were selling fancy watches from France, your customer support and sales team could have a French accent, just for fun. So it could really be used in a tonne of different scenarios and use cases.”

The technology can also be used in conjunction with popular collaboration tools such as Zoom or Hangouts.

Soderi says it currently takes a couple of weeks to create an AI model for each new accent. While the process of accent identification is still manual, the goal is to make this automated. The company’s mission will be assisted by an injection of US$5.5 million in seed funding from investors including Human Capital, General Catalyst, Quiet Capital and DN Capital.

Early tests have indicated up to a 40 per cent improvement in understanding when the technology is deployed between two speakers from different backgrounds. Sanas will be tested across seven BPO contact centres throughout the remainder of this year, and Soderi says the company has ambitions to see the technology incorporated in as many scenarios as possible.

“The end goal of the company is to help the world understand and be understood,” Soderi says. “We want to see this technology go into customer support but also healthcare, telemedicine, education or entertainment.

Source :