All News | Boards | Chips | Devices | Software | Archive | About | Contact | Subscribe
Follow LinuxGizmos:
Twitter Facebook Pinterest RSS feed
*   get email updates   *

Open source voice assistant speaker promises user privacy

Jan 26, 2018 — by Eric Brown 1,699 views

[Updated: Jan. 28] — Mycroft has Kickstarted an open source “Mycroft Mark II” smart speaker and voice assistant that runs Linux on a quad-core Xilinx SoC, and offers a 6-mic beamforming array, 10W speaker, 4-inch touchscreen, and a promise of user privacy.

When Mycroft launched its Kickstarter campaign for the original, voice-activated Mycroft home automation hub back in Aug. 2015, the Amazon Echo speaker and its Alexa voice agent had made a splash, but had yet to become a household fixture, and Google had yet to launch its Google Home with its Google Assistant agent. Now, the company has returned to Kickstarter to launch a more powerful, and similarly open source hardware and software Mycroft Mark II into a market in which sales of Alexa and Google Assistant based voice activated devices are soaring along with concerns about invasions of privacy.

Mycroft Mark II prototype showing calendar (left) and customizable robot face skills

Like the Echo and Home products, the Mycroft Mark II is a cloud-based platform. However, Mycroft is once again promising that user data will never be saved without a specific opt-in, and that it will never be used for marketing or advertising purposes. The free opt-in service is designed to improve the machine learning capabilities for greater user customization.

Mycroft Mark I

The Mycroft Mark II has soared past its $50K Kickstarter goal with packages starting at $99 for a build-your-own developers kit and $129 for the assembled device, both of which are due in Dec. 2018. This long lead-time may be one reason why Mycroft is vague on certain hardware details of a product that it claims will feature open source hardware and software.


The KS page never once mentions the word Linux, and says only that the device will run on a quad-core Xilinx processor. In the comments section, Mycroft developer Joshua Montgomery hints that the desired platform will be Ubuntu 16.04. His wish list also drops two more tidbits that are missing from the official spec list: 4GB RAM and 80GB storage. However, in an email to LinuxGizmos, a Mycroft spokesperson stated, “we have not yet confirmed which OS it will run, nor how much space it will have.”

Mycroft Mark II showing recipe skill
(click image to enlarge)

Assuming these specs are correct, and that the quad-core Xilinx is, as we suspect, the FPGA-enabled, quad Cortex-A53 Zynq UltraScale+ MPSoC, this is already a vast improvement over the original Mycroft, which ran Ubuntu Core on a quad -A7 Raspberry Pi 2 with 1GB RAM. The 16nm FinFet fabricated Zynq UltraScale+ MPSoC features floating-point DSP blocks, dual Cortex-R5 MCUs for improved real-time processing, and gigahertz-speed FPGA fabric.

Depending on the UltraScale+ model chosen, FPGA capabilities are likely much higher than the previous Zynq-7000 platform, with up to 914K logic cells. Presumably, the FPGA is being deployed to drive the new mic array, but it may also be used to assist local voice processing to offload cloud analytics.

Mycroft Mark II weather and timer skills

With the help of the faster processor, the Mark II can hear better thanks to a far-field, 6-microphone array from Aaware that appears to be similar that provided by the Echo. The array offers beamforming for better speaker isolation and active noise and echo cancellation.

The Mark II also offers an improved, 10-watt speaker with dual two-inch, full-range drivers. Together with the “ported and dampened sound chamber,” the speaker enables “vivid highs and deep lows,” says Mycroft.

The other major improvement is a new 4-inch IPS LCD touchscreen, which dominates the front of the 196mm (height) by 105mm (width) device like a vertically embedded smartphone display. By comparison, the earlier 152 x 150mm device had a horizontal configuration with a 32 x 8 pixel white LED array.

The new screen can augment voice responses for skills like timers, calendars, and weather forecasts, says Mycroft. It can also display “several friendly robot faces,” says the company.

Like the original, the Mycroft Mark II provides WiFi and Bluetooth (input only). There’s no mention of the previous Ethernet port, but new features are said to include a 3.5mm audio jack, a full-sized USB port, and an accessible microSD slot. There’s also an 18W power supply with adapters that support US, EU, UK, and AU systems. The device will feature FCC and CE certification.

Diagram showing Mycroft Mark II mic array (top left) and speakers (middle left), and on the right: a Mycroft chart showing open source voice technologies compared to other voice-activated platforms
(click images to enlarge)

More than 140 skills are available now, ported from the original platform, including creating shopping lists, playing music, reporting news, and telling jokes. Skills related to major brands include Roku, NPR, Twitter, Pandora, YouTube, Gmail, Facebook, Wikipedia, DuckDuckGo, WolframAlpha, and OpenWeatherMap.

More of these Python-built skills are on the way both from Mycroft and the larger Mycroft open source community. The skills are said to be backward compatible with the original Pi-based Mycroft, except for a few features, primarily related to the new screen display.

There appears to be a reduced focus on home automation. However, the device presumably continues the Mycroft Mark I capability of processing IFTTT logic and brightening or dimming lights on command. Automation skills include Wink and Philips Hue, and the video mentions the ability to display feeds from surveillance cameras.

Open source voice technologies include PocketSphinx for wake-words, with plans to move to Precise, and Mozilla DeepSpeech for speech-to-text. The system uses Adapt and Padatious for natural language processing, and Mimic for speech-to-text. There are plans to move to Mimic 2, which Montgomery says will enable more voice types.

The voice agent will initially offer a choice of American female or British male voices, with more to come. The developers are working on Spanish, Portuguese, Italian, French, and German support. Although Mycroft has left out a number of hardware details, the KS page provides much more background on the voice platform and privacy rules.

Further information

The Mycroft Mark II is available on Kickstarter for $99 for a build-your-own developers kit and $129 for the assembled device, both of which are due in Dec. 2018. Volume discounts are also available. The campaign runs through Feb. 24. More information may be found on the Mycroft Mark II Kickstarter page.

(advertise here)

Print Friendly, PDF & Email

Please comment here...