Closed StuartIanNaylor closed 4 years ago
Opensource design isn't easy because the materials and purchase qty you have often means everything is a compromise. With audio enclosures this is a problem as one thing you don't want to do is compromise in density and sound handling or have any components that maybe prone to resonate.
So here is my thoughts on the fundemental design.
1... Basic fundemental audio properties One of my first thoughts when I saw the design prototype was wow that 2 speakers on either side of a small device is not stereo as its just anti-phase which is actually what we do digitally with Mycroft to cancel noise. You can not make stereo like that as it just becomes 2 speakers in a singular sound source all it has done is place more constraint that speaker choice size has halved because we now have 2. Also its likely to be extremely annoying that because of the anti-phase 90 degree to listener orientation the percieved volume for listener is going to greatly attenuated whilst noise polution greatly increased. So coupled with the material density and component resonation I couldn't help think for a device in its price range its likely to sound truly awful in comparison.
If you are 3d printing I have seen dual wall XXX corragated designs that really help with the ridgidity and density requirements.
---
XXX
---
So its not XXX porn audio just a double wall construction where the X adds rigidity and also an angle of deflection in an almost similar way acoustic panels do. You can get fillers that help with the material properties also. If you are doing 3D printing one thing you can do that isn't really possible with traditional materials is make and design a tuned bass baffle. But much with sound is cone size especially bass, but look at Google & Alexa designs as these have not just been cost considerations but audio property decisions.
There is also one thing Google & Amazon do that I think is due to designing around a table is a 360 degree design for both mic field and outputed audio that is probably less common in domestic placement and a more directional design could give better volume and clarity. Thats a minor consideration though to the basic prototype audio flaws.
2... Engineering material choice. I started wondering about the dificulties in cost, time and final finish of 3D printing and if there are any alternative engineering materials that could give reasonable density, finnish, work and cost. Unless you have the production qty to merit injection molding then options are extremely limited. It did occur to me that schedule40 pipe laser cut into sections, with cutout options for display and electronics compartment could be valid as its possible to get low production runs laser cut.
Internally in compression there could be a stack of 3D printed compartments for speaker, CPU/ancillary electronics and Mic array. So basically you just building up a stack of components in tube shell that can be vinyl wrapped or even clear transparent acrylic for those proud of the internals Mycroft contains. Its the only alternative of a readily available preformed engineering material with good density, ridgidity, surface finish that laser cutting on low runs could still retain good cost.
But probably going off track but started thinking Mycroft could supply some quite high quality enclosure solutions to a market that seems devoid of any solutions. Drain pipe caused me some amusement at first, but the more I think about it, its actually just a really good engineering material. So its a choice of 3D printing or a tube segment enclosing 3D printing components with some consideration to double wall design for rigidity and vibration isolation. But in terms of basic shape the puck style tube has a lot of advantages as speakers are generally round and maximise cone size and go along with the 'pixel ring' style designs with have come acustom to. 4" is possible but some items are a tight squeeze, 5" makes thing much easier, 6" turning the project into an oversized lump. Still undecided on what should be AIX format but much is about the speaker.
3... Speaker compartment A tuned designed audio compartment can massively increase tonal quality and clarity.
I think probably your speaker is likely to be .5" to an inch less in diameter than the internal pipe dimention, but you could squueeze into the max if you choose.
Google home has done some quite neat engineering where they have created another cone shape facing the speaker cone that helps push the audio out into a 360 omnidirectional pattern. There speakers are top facing but I did wonder if a similar bottom facing might actually also gain some of the benefits of a vibration speaker and use that surface to enhance volume and tone. Doesn't work on soft furnishings but the tables and shelves these things usually reside it could actually work. I am not talking full blown vibration speaker same as Google home where a rubberised base with a cone on top will face the speaker that is attached to the speaker fixings on pillars to give approx a 10mm gap.
The advantage is that its not directly underneath the mic array for both pickup, assembly and center of gravity and the base actually becomes a functional property. I could be being hypocritical here as this might just induce resonance as it does create placement consideration, my audio engineering is lacking but if it works it does have some benefits.
Top or bottom we also have the option to add a tuned bass reflex port with a compressed S design that top vents to bottom or vice versa on whatever you choose.
The bottom side of CPU compartment forms the Speaker roof if bottom mounted or an oppisite arrangement if top mounted. The speaker compartment is just 2 directional cones with the possibility of a bass refex port that some simple adapter rings could give choice over cone size. If the Google and Amazon is design around a table for a device that really needs to throw forwards, that can also be incorporated, dunno but I would go for unidirectional rather than omnidirectional in use. MIcs it makes sense for AEC but audio output less so on how they are often placed.
4... CPU/AMP/Mic compartment
Depending on chosen speaker orientation the base or top seals the speaker compartment, as said. Really all it is a cutout with a blanking cover that allows a blank FRB board that is screwed to the compartment base that you can drill and attach an array of different SBC and amplifier boards to. Basically you use and array of pillars and create a stack of SBC/AMP/Mic depending on BoM used.
The blanking cover is an open box shape just to allow an extra inch of length for a 4" enclosure. Its 60mm high has a 40mm extractor fan on it, 'DC in' above the usual Pi cutouts that may have a options for Pi3, Pi4 and blank for a DIY. Also has a USB cut-out for a 180 degree usb connector for LCD/Mic, dunno really do you supply premade or just a blank that is a matter of a bit of DiY and a dremmel? The actual lid is just an Acrylic disk that again sits on a pillar arrangement irrespective of respeaker2, mic array, Pi or amplifier there is space and positions where these can all fit. Its just one of those printout option drill guide jobs where someone else has worked out optimal spacing for common components and the option for users to DiY there own. Just needs vinyl cutouts for the various mic arrays to cover and hide the internals but give the LED windows and mic holes again a print out drill guide. I guess also instead of just a Acrylic disk you can get posh textile dome of some sort and fit if it was made available.
4" fits up to Pi format boards and Re-speaker Core2, Pine64pro and jeston wouldn't fit due to profile and heatsink. 4" will fit up to Pi format boards and respeaker-core 2 which gives a lot of options without being massively oversized, so I am thinking rather than 6" not fitting a Pine64Pro or Jetson is just a compromise to make.
Until you have all things infront of you its not good to give definates but its should all fit in product height of approx 130-150mm
5... LCD monitor & Camera
The 4.3" display is only 2.5" wide so it is really possible still to create a cut shaped blanking plate that it would hold, just shaped to fit on the outside 4.5" tube diameter. The curvature gives only aprox 5mm difference so it would protrude but still fairly minimally The round shape does lend itself to posibilly panning the whole
Also a flat fill blanking plate to regain a filled smooth surface for those who may not desire one. Or go flexible e-ink but the refresh is quite ropey of .5Hz but it would be quite cool with a wrap around display
Camera is interesting as the pan could be the whole unit with a swivel base and maybe power should go to a slip ring and maybe the speaker should be up top. But the display would always face you and Robot Myford would be quite unique, but it would demonstrate the DoA (Direction of arrival) capability that is in place.
Haven't given the Camera to much thought apart from power via a slip ring but the main importance is this can all be modular and sold as kits with a far wider market than the very specific prototype.
6.. Conclusion.
I wouldn't buy a Myford II IE the pan-tilt complete camera/ display Myford as I am not just that type of tech geek guy. I have a feeling though initially I would of been really happy to get any kind of enclosure and just purchased a kit to house a raspberry Pi and Mic array and run off blue tooth. Bluetooth A2DP highquality audio works great in Linux it just seems HFP and mic usage is the only problem. So might even not bother with a speaker or amplifier. I prob would at one stage out of interest as Myford could also be a bluetooth speaker and the amp/speaker I had in mind is actually not bad quality. I can imagine myself being intrigued and robot Myford DoA (Hopefully not the dead sort) and buying the pan/tile kit and a Pi camera. It would then be an oh sod it lets get the display also. So not being that type of tech geek guy I can't imagine buying a Myford II but probably I would end up with one.
I couldn't imagine myself buying one of your protypes either and just wouldn't because its a specific piece of kit that has an inital price hit that makes me think of more sensible options. Probably if it was more modular I would likely get the Myford builders and curiosity bug and keep adding in stages. But there people out there with Respeaker Core2 and other equipment that might not be interested in Myford but just want the enclosure / speaker kit. What I have suggested isn't really any better, but its modular and has markets rather than a specific niche.
I really do think you need to make it modular as its likely you will sell far more. Also sorts a patent troll problem out instantly, build kits guys.
Hey Stuart, just closing this as the discussion continued on the Forums: https://community.mycroft.ai/t/hardware-recomendations/8251/7 and didn't want confusion with the new designs we just released
Would be great to get your thoughts on the new prototype. I believe it already incorporates some of your comments: https://github.com/MycroftAI/hardware-mycroft-mark-II/tree/master/mark-II-Rpi-devkit
The Pi4 2gb recently dropped in price as the 1gb model is no longer produced. https://cpc.farnell.com/raspberry-pi/rpi4-modbp-2gb-bulk/raspberry-pi-4-model-b-2gb-bulk/dp/SC15186 https://cpc.farnell.com/raspberry-pi/rpi3-modbp/raspberry-pi-3-model-b/dp/SC14882 Strangely at farnell the practically same price but the difference is so minimal. Pi4 you do need to run with a heatsink so guess that adds cost. https://cpc.farnell.com/pimoroni/com3100/aluminium-heatsink-case-for-pi/dp/SC15306
But its actually a big upgrade that creates much load potential.
This is why I am bemused as the relatively small increase in pi3 to 4 has been chosen whilst the speaker array. https://www.seeedstudio.com/ReSpeaker-Mic-Array-v2-0.html is much more than https://www.seeedstudio.com/ReSpeaker-4-Mic-Array-for-Raspberry-Pi.html Maybe the choice was because of
But the Pi4 is more than capable and the ODAS engine https://wiki.voicen.io/audio_processing/doa/ is sort of cool. VAD is also easy and relatively low load. Much with the above choice is using closed source Alango routines whilst we would go totally open-source and also knock $40 off the cost and that means with the Pi4 upgrade and heatsink we are still $25 cheaper.
Alsa when it comes to cost of using the latest audio AI technologies the touchscreen is really superflous whats great about Mycroft is the huge number of skills. To have the time displayed whilst being told the time is nice but actually not nessacary, I am certainly not going to be squinting at a 4"3 screen and watching video. But to have it built in is sort of contary to the nature of IoT and even if a Android app is too much to develop a virtual display and any phone could do the same function via VNC. There are much cheaper offerings personally the screen is as said and reduces a lot of cost if omitted but maybe shop around and go for 7" portrait? But still in its format and likely positioning the screen could be optional and maybe that should be part of the enclosure design with a fitted cover / blanking port? If its going to have a screen then not having a camera is an omission.
Also why have I2C volume when the master volume control of the input will do the same? I purchased a 30watt mono amp as at 24v DC 4ohm you can fit a much better singular speaker. Stereo in such a small device is relatively pointless because there is no seperation in reality. I got an aliexpress cheapo that actually sounds great https://www.aliexpress.com/item/32848281225.html
Also a USB hub(s) on all the ports we are stealing would not add much to cost.
I quite like the enclosure but the selection of components as a personal opinion would seem to be unessacary and missmatched. I would love to get that ODAS lib into Mycroft and use a PI4 for the more load scope it gives the project and add BT5 and disable the broadcom internal.
PI3 with mycroft running is quite load bound PI4 opens up some interesting options of https://www.domoticz.com/ and spare USB ports for zigbee/zwave and even mirocast.
Needs to be much more modular with a base system that throws much out without much loss from what I see of the current BoM. There is a hole in the maket for a Pi/far-field mic/speaker case in its own right and then add mycroft to that. When you have an open and module SDK like mycroft that is interchangable with various framewoks and libs then why not have an enclosure system that is in the similar vain?
Just thought I would fire that one across your bows not in a negative, but just wondering you not missing an oppurtunity when there isn't a enclosure for most of the AI kit available. Continued this on the forum https://community.mycroft.ai/t/hardware-recomendations/8251/7