'Sirius' Is the Google-Backed Open Source Siri

Sirius wants to be the Linux of virtual assistants.

Mar 11 2015, 5:16pm

​Image: Joseph Xu

​Sirius, a Google-funded open source program similar to Apple's Siri or Google's Google Now voice recognition application, could finally democratize the virtual assistant.

RIght now, virtual assistants are a game for the big kids of the tech world—Apple, Microsoft, Amazon, and Google itself all have their own versions. Sirius, developed by researchers at the University of Michigan's Clarity Lab, aims to do what those programs can with an open source twist.

Other backers include the Defense Advanced Research Projects Agency, the US military's research wing, and the National Science Foundation.

The idea is that anybody can contribute to the program on GitHub, a site for coders to collaborate. It's also being released under a BSD license, documents on the project's GitHub indicate, meaning that it will be completely free for anyone to use or distribute. Researchers will be able to use it to explore the possibilities of virtual assistants, according to a university statement, and eventually, anybody can put it on their own homebrew device.

Right now, it's only been tested on Ubuntu desktops, but it could one day make it onto phones and other devices. Jason Mars, the researcher that headed up the project, describes Sirius as a Linux-like version of Siri.

Sirius already has capabilities lacking from its corporate counterparts. For example, you can take a picture, feed it to Sirius, and ask a question about it. Siri can't do that. But, unlike Siri, Sirius isn't exactly elegant; it's a patchwork of other open source projects that, when stitched together, give Sirius its capabilities.

Naming all of these programs is a bit of a mouthful. To recognize speech, Sirius uses Carnegie Mellon University's Sphinx program in combination with Caffe, an open source deep neural network platform. To understand images, the Sirius uses OpenCV's SURF program and image database, which IBM's Watson also uses. To actually answer questions, Sirius uses a system called OpenEphyra, also developed by researchers at Carnegie Mellon.

Credit: Mars et. al. 

You'll have to download all these programs to make Sirius work, but thankfully they're all available as part of a downland suite. But downloading the suite and actually getting the components to all work together are two very different things. Thankfully, the team behind Sirius is running a tutorial on how to work the damn thing, as well as ​presenting a paper, next week at the 2015 ASPLOS conference for programming languages.

Other programmers have tried to make an open source version of Siri and failed—the GitHub page for JuliusJs, a project that took a similar open source patchwork approach, shows that the last contributions to the project were made 8 months ago.

Will Sirius succeed where others have failed? For now, that's one question that it can't answer for itself.