This Google-backed startup is creating a Siri-like voice assistant for companies like Spotify
On Thursday, MindMeld (formerly known as Expect Labs), a San Francisco-based voice technology startup backed by Google, Samsung, and Intel, announced it is launching a language understanding and question-answering system that can be customized to any type of content.
"We're trying to make it possible for us to talk to our machines, like in science fiction movies," company CEO Tim Tuttle told Business Insider.
The music streaming service Spotify is one of the first companies to use the new system, known as MindMeld 2.0.
MindMeld claims its voice system surpasses the performance of other intelligent assistants like Siri, Cortana, and Amazon Echo on these custom data applications.
Until recently, voice recognition software wasn't very good. But over the past year, there's been a huge uptick in the use of voice assistants like Siri and Google Now. According to MindMeld, as much as 60% of smartphone users use voice search, and more than 80% adopted this feature in the last year. And some experts predict that 50% of web searches by 2020 will be done with voice (or images).
"MindMeld is a platform that can help every company build an intelligent assistant for its own applications and own data," Tuttle said.
Spotify is one among the first to adopt MindMeld's platform, to help its users search for and play music by voice. MindMeld says its other customers include some of the largest cable and car companies, though it didn't specify which ones.
The company released a video comparing its voice platform to Apple TV. While both systems perform well on basic queries like "Show me Harry Potter movies" or "Which Steven Spielberg movies star Harrison Ford?", MindMeld appears to perform better at questions like "What are the Oscar-winning movies of 2014?"
In June, the company partnered with Sense.ly, a "virtual nurse" application that provides personalized medical monitoring and follow-up care, to give users the ability to ask questions about symptoms or treatment options.
Like other voice recognition technology, MindMeld's platform relies on large-scale AI techniques such as deep learning, which learns by training itself on enormous datasets. The company's software focuses on understanding speech and question-answering technology similar to IBM's Watson.
"The way these systems become good is they have models that are trained on actual usage data of how people use these applications," Tuttle said.
The company is offering both a cloud-based version of the voice system, as well as a hardware version that can run on a customer's own network.
Watch a demo of MindMeld works here: