Modifying Voice PE

[vestal hull]

Curious if anyone has modified their Voice PE hardware. Been curious about buying a cheap echo dot and changing out the internal speaker and microphone on the PE for the ones from the Dot.

[rotund hazel]

so... first question... why? you want to sneakily replace a device under someone's nose without them noticing?

but do the question at hand

the speaker is connected to a header and could easily be changed. the microphones however are fixed and connected through a buffer to the XMOS chip. now... could you still resolder the mics and connect others... yes but its a bigger job than just plugging in a different microphone. also i beleive that the XMOS chip firmware is calibrated for the built in microphones and their distance apart etc... to do noise filtering. so you may need to customise that chips firmware. then have the VPE firmware customised to not overwrite it.

from a realistic point of view, if you can do the above you are likely better off just making your own custom circuit. if you skipped the xmos chip then you could get the majority of functionality with a few components. I built a custom satellite inside an old pc speaker a while back.

btw: voice PE schematic can be found HERE

[vestal hull]

what are you talking about? Sneakily replace a device? It's my HA Voice. The Speaker on the PE sucks, and the microphone isnt great either. So Why would you not want to transplant new equipment intoi it if given the chance?

For the other stuff... Yea i figued it was going to be harder than just desolder and solder sadly.

Also thanks for the schematics!

[rotund hazel]

yeah the speaker is only really good for voice and even then depends on the room. for a bigger room i used the 3.5mm jack to add a better speaker that sits next to it. WRITE UP HERE and it works great. when i get a 3d printer at some point I will probably make a little enclosure for them both.

i dont have any issues with the microphone. if you want something a bit easier to make a custom satellite with then you should look at some of the ReSpeaker boards. they may be useful to you. although not sure if they would hit inside a dot.

another option is to use a RPI (or simalar) and then you could use THIS newer project that way you could use any speaker or microphone that you want and can connect to the RPI

[vestal hull]

Oh thanks! Actually super helpful! Actually may just go that route. It wasnt the speaker is bad, its just i am so used to my 1st gen echo dot and being able to quietly (normal indoor voice) wake alexa whereas HA PE i find myself shouting to wake it (same distance as echo (10-15ft away)).

I ended up buying a used echo Dot 4 since that seems to have a nice condensed speaker design in it. Will look into that linux HA Voice repo and a respeaker and maybe 3d print a new case to hold everything together

[rotund hazel]

you can adjust the wake word sensitivity and also try other wake words to see if you can get it more smooth. but yes the wake word detection is not perfect.

[unkempt mural]

My PE goes unused because the wakewords are extremly poor for everyone in this house. “Okay Nabu” gives far too many false negatives. The others too many false positives. It’s our southern England accents I guess. But yup it’s lame. Never had any troubles with the Amazon products in this regard. And yeah the speaker is miserable beyond words, though I mostly expected that. I really hope Nabu sort out the wakeword problems and also make a more larger unit with a decent speaker but unfortunately I’m not particularly hopeful about either thing. In the meantime my PE is muted to stop false positives and makes a nice speaker for notifications.

[vestal hull]

I mean... i have it set the highest itll go. But just not that great with hearing. Actually just bought today the ReSpeaker XMOS XVF3800. 4 mic array setup running an esp32 so im going to take that and jerryrig that into the Echo dot speakers and build a case for it

[rotund hazel]

part of the issue is having data to use to improve. you should help out by adding some voice samples to the OHF Wake Word Collective whilst thousands of people have done this its still a little limited, unlike amazon who when they want data they just slerp millions of peoples data without asking.

[vestal hull]

yea i was looking into just finding a drop in replacement but its only like a 1w speaker so only so much you can do without going 3.5mm jack

[unkempt mural]

I bought mine a month after they were released and as soon as I saw m this problem, which many people don’t I happily admit, I submitted my samples and hoped for a new software release. I also found out that I wasn’t alone and many native English speakers who aren’t also American have this trouble with the PE. Of course the userbase is tiny compared to Amazon’s. But the lack of software updates is still disheartening as is an acknowledgment of the problem.