Mycroft Open Source Voice Assistant
Voice activated assistants are everywhere. Capable of handling many simple spoken requests, they make an ideal partner when creating shopping lists, checking the weather or reviewing our calendars. However, many of us remain cautious about letting a microphone, which theoretically is capable of recording intimate details of our lives, into our homes.
None of the existing voice assistants, such as Amazon’s Alexa or Apple’s Siri, are capable of decoding complete human sentences. At most they can recognize the wake word, such as ‘Hey Siri…’, but not the remainder of the command. The remaining audio has to be uploaded to an AI-based cloud service that converts the speech into text. The text is then decoded for meaning, allowing the desired command to be actioned.
This means that the service provider could garner insights into our purchasing habits, our favorite TV shows, how much time we spend at home, and ultimately use this to deliver targeted advertising through their other services.
Now undertaking their second round of Kickstarter funding, the team at Mycroft have taken a significantly different approach. In 2015, the team successfully funded the Mycroft Mark I, a collection of hardware and software that hackers and early adopters could build into a voice assistant akin to an Alexa or Echo from Amazon. Building upon that success, the team is now busy developing the Mycroft Mark II.
The Mark II has a significantly upgraded hardware. The brain behind the device is a Xilinx quad-core processor that is coupled with a color display, punchy speakers and an array of six microphones. This is built into a stylish case with a cloth cover.
Where the Mycroft solution stands out, is in its approach to software. Firstly, the entire software project is open source. This means that anyone can examine the software and check it for potential problems. The main area of concern is typically security, meaning that the community is both capable of researching for weaknesses as well as fixing them – an approach Mycroft’s competitors don’t allow.
Next, it allows a community of developers and users to extend the software, allowing Mycroft to potentially interact with any number of web services, Internet-of-Things (IoT) devices or Smart Home appliances.
Privacy is also ensured in the portion of the functionality that is handled in the cloud. Mycroft uploads the audio to be converted into text for analysis but, upon completion, automatically deletes that data. Only through an opt-in process can such audio data be retained by Mycroft. Even then it is only used to further train the speech-to-text AI algorithms for better detection of different dialects and pronunciation.
With privacy concerns being highlighted regularly in the media, and worries surrounding how the big tech firms use and share our data, Mycroft seems to have the correct solution at the right moment.