I want to build a Picroft for testing and if I like it, i want to build one for every room. My main goal is to control my smarthome devices by voice. Well, I mostly want to control openHAB with Picroft.
I also want to build them in different shapes, like a ceiling cube, that includes Picroft, some beefy speakers and smart room lighting…
But for now I’m still wondering which mic array to get. As I see it, there are three (circular) Respeaker Arrays to choose from:
ReSpeaker 4-Mic Array for Raspberry Pi (25 bucks)
ReSpeaker 6-Mic Circular Array (40 bucks)
ReSpeaker Mic Array v2.0 with XVF-3000 (70 bucks)
So the million dollar question is, which ones the best? I really like that they all got RGB LED. I think the cheaper two are not very different, except that one uses 4 and the other uses 6 mics.
What really interests me, is how much better the hardware accelerated one with the XVF-3000 chip really is, is it worth the premium I would pay? Did anyone of you compare it with one of the cheaper options?
Advantages of multi-mic arrays are echo-cancellation and beamforming for better voice-recognition.
You can do echo-cancellation with the 4-mic linear array and the 6-mic array (the 2-mic and 4-mic can’t do EC as they have no hardware loopback). All versions with with at least 4 mics can do beamforming, but the cheaper arrays without hardware acceleration require additional configuration in the audio subsystem of your operation system.
When you want to use these mic-arrays with Picroft you must the Debian-Stretch version (which is currently still “unstable”) as only this support Pulseaudio 9.0 with the proper echo-cancellation algorithms and beamforming support.
Both the 4-mic array as well as the 6-mic array (both circular) uses the same driver and work perfectly. I have the 4-Mic circular array, but will eventually switch to the 6-mic for the reasons @Dominik already hightlighted. (6-mic came out just a few weeks after I bought the 4-mic. Doh!)
They pick up “hey mycroft” from at least 25 ft away, around the corner of a wall, with a normal speaking voice.
I have been trying to get my respeaker 2-mic phat to work, with some success. It required installing the drivers from the seeed github, editing the mycroft.conf file to use the correct card in aplay and mpg123, and setting the default soundcards in /etc/asound.conf. Then for some reason after the pi boots I need to run ~/mycroft-core/start-mycroft.sh voice and reconnect to mycroft-cli-client before mycroft will listen for the wake word. I haven’t yet figured out why the last step is necessary.
I postponed my whole picroft project for now. After I learned that STT is done on HOME and not locally it is no longer an option for me.
There is some talk here on the forums about a potential personal mycroft server. As soon as this becomes reality, I will be back, because I really want to voice control my home. But I will not have any cloud listening in my home.