Hardware Recomendations

I have use for another Mycroft unit in my home, but they’re a little hard to get a hold of these days. Mark I is out of stock, and Mark II isn’t expected this year. I want an all in one unit that encloses the speaker and microphone (no USB bits). I have a 3D printer, so I’m happy to print an enclosure. I don’t want to buy a Google AIY for Googly reasons. Does anybody have a list of hardware that just works with minimum interaction that fulfills those requirements? I’ve already checked the list found here, but nothing really jumps out as something that I can use to do this. Advice would be greatly appreciated. Thanks!

Have you considered the Respeaker Core V2?

During the Mark II protyping phase this was a candidate as SBC and a enclosure was designed for it: https://mycroft.ai/blog/mark-ii-update-revised-architecture/#phase-ii-optimized-audio-prototype

Mark II dev team promised to published the STL for this - but didn’t happen until now. Maybe they do if I am not the only one asking for it :wink:

1 Like

I suppose the tidy solution is the respeaker core2 which is a rk3288 which at a guestimate is somewhere between the pi3 & 4 in terms of performance.
You will prob just be running debian rather than raspbian.

But until the Mycroft II arives presume there is nothing that looks as finnished as the available cases and hex board arrangement that fit.

Bit pricey though but you get the snazzy far field mic array and the Alango closed source audio processing.
Strangely or at least I think so they didn’t give room for a speaker but you can BT5 that.

I am a big fan of the Pi4 as the 2Gb has more oomf and also a great price but from googling myself the enclosure is DIY.

The Respeaker as @Dominik said is quite handy though as they include mycroft in there wiki and also you can use it as a fancy cutting edge far-field BT5 mic and Mycroft could reside anywhere.
Mycroft-Cardboard :slight_smile:


1 Like

The CoreV2 does not fit into the “Respeaker Pro Case” - I learnt it the hard way…

1 Like

So its version 1 only then ?
That pretty crazy innit :slight_smile:

Get the google aiy kit v1. No google involved if you flash with picroft.

1 Like

I seem to be of a similar opinion of the OP @linuxrants that its actually quite desirable to have something that resembles a ‘finished’ product that is not just about asthetics but a robust working unit.

It really bodes well for Mycroft that respeaker core 2 doesn’t have an enclosure off the shelf.
Its also why I think the Mycroft II prototype is far off the mark as there is a huge array of kit(s) crying out for a simple enclosure system.

I am not at all sure why its has such a specific BoM and design rather than a modular system that can be an open extensible system of many options and for many reason shouldn’t be a specifc model.
That it could be a Respeaker Core2, Raspberry Pi to Jetson Nano and really suggest thinking about some simple building block solutions that can complete an array of options to provide finished products that are fit for purpose.

I think you can have a series of stackable ‘disk’ compartment modules that sit on top of each other and are interchangeable to give a huge array of options.
If you design it well you give even set a typle of AI format that AIX could be the ATX of AI.

Likely needs to try and gain a low center of gravity and get weight into bottom compartments and stack up with lighter units. But also sound advantage of density.

Mic array should have choice so it suits a wide array of taste, pocket and application.

Starting with the excellent value 2 mic https://www.seeedstudio.com/ReSpeaker-2-Mics-Pi-HAT.html
Its $9.90 and its amazing how hard it is to get a multichannel mics as all equivalent priced USB audio adapters seem to be mono.

I have been slowly chipping away at pulseaudio webrtc beamforming module and yesturday found a DoA (Direction of Arrival) lib as its been puzzling me how does it beamforming with the DoA that seems missing from the code.

GitHub - voice-engine/voice-engine: building blocks to create voice interface applications has a DoA wake-word routine that finally allows me to add the target_direction to the beamforming routine of the WebRTC AEC module.
Its rather inelegant as I am thinking I will have to pactl unload/load the module to update the target_direction co-ordinates but hey this is a base cheap and cheerful option.
Also can be fitted with a adafruit pixel ring but again its extensible from a very basic base.

The 4mic array is excellent value has DoA and all the goodies with an inbuilt pixel ring.
Hard to beat @ $24.90

Then the step up to the ReSpeaker Mic Array v2.0 - Seeed Studio
$64.00! but that XMOS audio processor is cutting edge.

If the top of the stack (which it can be or not, but will get to that later) is a single compartment that stacks on the audio / psu / hub compartment with and an array of fixing holes that fits all or they are simply interchangable compartments of a common format.

Then the next in the stack is the amp / buck / hub compartment.

Next is the SBC compartment as with heatsinks they can form much of the low center of gravity.

Bottom compartment is for speaker(s) and may not even have one as BT5 and true satelite seperation might also be a choice option.

Each compartment is a series of open frame stacks that fit into a PVC or acrylic tube that can be vinyl or cloth wrapped to finish.

That further options should be LCD displays but here again there should be a huge array of choice.
From the adafruit ‘Eyes’ module

To flexible e ink or Oled

and even standard LCD touch screens if you wish.

I dunno what worldwide standard tubes are but 110mm as an example could squeeze but whatever is can be standardised uo to 160mm as at that size its AIX format but things definately fit.
The stackable compartments don’t need to be much more than fixing rings that will help with aperture stacks if touch is going to be included but I have to say touch and far-field seem totally different applications to me.

There is loads to think about much thought needs to be about material density or Mycroft will sound like a plastic hollow box.

To crown the stack it could well be possible to provide a PT (Pan/Tilt) camera as all the above arrays have direction of arrival technology.
That this could be for out-of-ocupancy security, baby-monitor or inter-room to door video conferencing.

Its a niche market and making a specific niche product in a niche market narrows numbers massively and why it doesn’t make sense to create a specific BoM product when you could satify an array of options, budget and applications.

Mycroft could have instant possibilities in being the Adafruit of AI as its extremely frustrating for many that there is a lack of some simple enclosures, never mind an extensible AI format that offers an aray of choice quite simply.

1 Like

If finished = cheap hardware that is known to work well + 3d printed case, that allows for a great range of answers that qualify, so lmgtfy…

Turns out there’s quite a few interesting options already available. I like this one: Companion Cube - AIY by i3dKC - Thingiverse

1 Like

Haven’t used Google AIY but to be honest not a fan.

From playing with Pi3/4 load of wakeword/beamforming, STT , TTS of any quality its not going to work on a Pi Zero.

Even Magicmirror which is liitle more than a very simple local webserver/webapp recommend something more than a Pi zero.
Google AIY bemuses me as the fun is likely the build but definately not the use.

When people here mention Google-AIY they most likely think of the v1-Kit that runs with a RPI3. Don’t know if that is still available though.

Dunno as its not really what it runs on, its what the SDK can do. Have never bothered with it and it might be a presumption its extremely stripped down with no echo-cancellation, beamforming…
Just presumed so as it can run on a pi-zero and my logic is it much can not exist as there isn’t a hope on the zero.

Prob assumption but it looks like a basic tutorial kit and introduction than anything of any use?

Well I can’t wait to see the writeup of your ideas.

1 Like

Already have its in the above Hardware Recomendations

Its purely a matter of choice as there is some great kit available, but just requires a simple enclosure system that can stack.

Its that simple it doesn’t really warrant a writeup.

Also commenting on https://github.com/MycroftAI/hardware-mycroft-mark-II-rpi/issues/2
Purely feedback but thinking actually Mycroft could fill some gaps in the market that have much wider turnover.
The mycroft-mark-II is cool but its very specific in what you can use with it and just wondering why?

1 Like

Noob question: is a USB speaker/microphone possible?

I’m ramping up to implement Mycroft. Was conceptualizing using a Linux box I have available paired with something like a UE Boom speaker/microphone.


Yeah but like the lower end Respeaker stuff have a look at what they do with vad/aec and stuff as only the higher end have all those goodies and libs thrown in.
So you can do what they are doinf with the lower end stuff.
You should be able to use bluetooth as long as you stay away from HSP and do it a2dp.

yes, most people use usb mics and speakers.

I couldn’t find much about the UE Boom in terms of microphone apart from its dual.
Much about the UE Boom is that its a high quality BlueTooth device where you can ‘double up’ and play a pair as stereo.

Much about those speakers is the buetooth wideband audio and the ability to link 2 in stereo and they will latency sync.
BT5 has wireless beamforming built in and I think they use this so actually its inbuilt BT5 functionality but not absolute sure how these devices sync.
Not exactly sure as maybe it does use its stereo microphone in the same way Mycroft uses multi mic far field arrays.

I am actually awaiting delivery of a BT5 speaker mic and if that goes well will be buying another as I am not sure how well all this goes together in reality.
In BT5.1 they introduced Angle of Arrival (AoA) and Angle of Departure (AoD) which are used for location and tracking of devices
I am presuming this is what is used in the high end BT stereo syncing speakers rather than mic arrays.
In BT5 they doubled the bandwidth and quadrupled transmission difference so now you can run 2 wideband high quality A2DP audio streams concurrently.
It will also auto switch between to hosts and all whole load more features.

But before you go out and splash £100+ and more on a Blue 5 speaker mic just bear in mind there are under £10 dongles that do the same.


You could just stick a dongle in a dumb powered satelite speaker.
But also theoretically and what I am working on is that with BT 5.0 you can have Myford and a single dumb BT5.0 speaker work as a stereo pair.

There is so much with BT5.0 that my head is still spinning and will not have a full picture until I actually test.


But generally BT5.0 is a game changer for audio distribution and home automation, but where BlueZ, codec support and function comes in is confusing.

If you do a trawl of youtube there are some really good videos and its surprising for $ what you can get.
I have a TPA3116 board which matches well with a 3.3" 30watt RMS speaker and a 24V supply.

The original UE boom was only 9 watt and they do so really odd s*** of supposed stereo that isn’t just audio snobery its just basic audio physics.

Its actually quite interesting what you can do, but beware there is an absolute plethora of false claims and overpriced trash available.

Its very dependent on what you want and what you have got but its likely a Bt5 audio reciever attached to your existing HiFi will blow the proverbials off what is currently a marketing festival of BS in personal audio.

I will be doing a write up of my findings of a 1+1 bluetooth system of a Myford & single satelite with matched cones and amps in stereo.

Just wish I has scopes and studio mics so I could get really geek audio engineer and present facts rather than opinion.
Dunno what the Dacs are like in those el cheapo dongles but at that price pron worth a go, do a google for a few reviews.

Also when in stereo mode if the mics are available or switched as stereo would be 2x a2dp wideband streams and the max? As it is still slightly confusing.
But again bluetooth with its frequency hopping can run multiples together with little interference.

And yes, as some of you properly interpreted, I had meant to type bluetooth speaker/microphone, not USB. Doh!

Its OK but as I often do the same :slight_smile:

Bluetooth is a bit more complex than you might think.
Needs to be a2dp for both mic & speaker as hsp/hfp is a nightmare.

I purchased a Flang as 5.1 devices cheap are pretty new and the cheap imports are often a Lotto.
If you already have a speaker then give it a whirl, but a cheaper option to try thing out might be something like that flang.

Is the best info I have found and even after reading it I am not sure, so hence I am doing a ‘suck it see’ trial.

I still don’t if you sync 2 speakers does that steal the mic bandwidth and it presumes its just stereo mode then.
Then there are the different versions and codecs…

Hence why I am going for a 1+1 that is my own construction so the cones are matched as an amp input of a BT reciever like the flang is fairly easy for me.

I also came across some new technology with speakers that will work much better and maybe those devices conatin similar.
Wave bending speakers is a new one to me and not that expensive.

The google units use smaller ones of those but you can read up about wave bending :slight_smile:

In Bluetooth 5.2 they released LE Audio as one of the reasons is to give a default format because currently there are a lot of proprietary adaptions.

There are lots of good ideas here, @StuartIanNaylor, I’ve seen your recommendations on github and in other parts of the forum. There are some really good suggestions I especially like the idea of making a modular system. I can expand a little on why the Raspberry Pi based prototype is the way it is now, and why it has a specific BOM (instead of something more modular). The biggest reason is that it wasn’t designed from the beginning to be a DIY or Maker project. The design was modified from a more production-ready design to accommodate some off the shelf components. The quick goal was to give our internal developers a dedicated piece of hardware, and also have something we could show potential investors as buyers. Although it wasn’t perfect for DIY we decided to release the design anyway. However, I would like to go back to the drawing board and work on a design that is meant to be a lot more accessible.