Microphone, camera integration, and noob questions ;)

[wibimaster]

Hi !

I’m ready to try my first attempt for a magic mirror :D

Got some noob questions :x

• I have a ReSpeaker Mic array v2 (http://wiki.seeedstudio.com/ReSpeaker_Mic_Array_v2.0/), where should I place it ? Behind the screen ? How it can correctly listen to me ? :/
• I want to add a camera (I think the standard Pi cam), should I place it behind the glass (colors/luminosity ?) or do I drill a little hole in the wooden outline to “let the sunshine in” ? ^^

Objective :

• MVA display (for good contrast and avoid light in the night) 24" or 27", wood structure, Mirropane 4mm glass

And at least, no laptop screen exists beyond 18", so why a laptop screen + some electrical stuff are so lightweight and small, and on a regular PC monitor, there are so much electrical equipment ? -.-’

Thanks for the pro !

[Fozi]

@wibimaster said in Microphone, camera integration, and noob questions ;):

MVA display

Well, to answer your question specifically, it might be good to know what you want to achieve with the mic. If you want to control the mirror ang e.g. switch pages, turn display on/off, then it is sufficient to put an Echo Dot behind the mirror, connect it to the RPi as power supply and use the MMM-AlexaControl module. This is the way I went and I am totally happy with it.
That is less expensive and totally hasselfree compared to installing a dedicated mic and installing and configureing a module that emulates an Echo device on the RPi. If you intend not to use Alexa services, e.g. Lucy, ten it is of course a different game.

The PiCam should work behind the glass. But maybe you do some research regarding red blured images and IR-filtering or automatic IR-cutoff provided by the cam. I made some experiments with a PiCam (15 € knock-off from AliExpress with automatic IR-cutoff, great cam that no is going to be used with OctoPi) and facial recognition but did’t succeed. The resources needed are pushing the RPi to the limits and reliability was zero. Further I found no real user story to implement that into my MM at the end.

The MVA panel combined with the Mirropane Chrome Spy is a good choice (also my combo), as it provides a pretty good contrast compared to TN panels.

[wibimaster]

Thank you !

I use Snips, not Alexa; that’s why I already have a ReSpeaker, but it’s like an Alexa device, circular and with directional microphones. Like Echo dot, it is intended to be used horizontally and not vertically… But if you said the results are good enough behind the mirror, I’m going with that :)

Maybe I could take a bigger mirror and let a space at the bottom to have the mic (and its LED) stuck close at the mirror ^^

For facial recognition, what hurts the Pi ? The PiCam itself (so an USB camera should be better ?) or the multiple call to photo / recognition / photo / recognition…?

[Fozi]

For facial recognition, what hurts the Pi ? The PiCam itself (so an USB camera should be better ?) or the multiple call to photo / recognition / photo / recognition…?

It iS not the cam that demands so much resources. It‘s the software and the algorithms that allocate RAM and CPU to accomplish recognition. It is of course amazing that technically a small RPi is able to do that and demonstrate the potential on HW and SW. But still, I sometimes ask myself, is it really beneficial and adds it so much value to the „customer“ to justify the effort and expense? I admit that this question shouldn’t be asked under us mirror builders ☺️ I‘d be happy to see how you accomplished that and what your experiences are, as I‘m eager to learn.

[wibimaster]

@Fozi For real time application, yep ; with an acceptable delay, recognition could be done online ^^

An offline example for good performance on RPI : https://medium.com/@aiotalabs/deep-neural-network-on-raspberry-pi-c287e06a3250

Would be happy to test that :D