Weird behaviour of VR3.1 module in circuit

[Moriambar]

Hi there.
I am starting debugging a circuit of mine, from last year. It features an Attiny804 controlling a couple of led strips and itself controlled either by buttons or a voice recognition module.
Basically a switch (J7) turns on or off a mosfet which, in turn, connects the "GND" line of the recognition module. The switch status is sensed by the tiny which executes code to enable the mic or the buttons.

Everything works fine and dandy as long as my PSU is powering the circuit. But the friend I built this for uses powerbanks to power it and, while buttons work fine, it turns out that the VR module does not.
Furthermore I tried to see the changes in current by using my dmm6500 to capture whether this could be a problem BUT with the multimeter in-circuit, while it clicked through the ranges, the VR module started working even powered from the PB.
To measure current I then tried and put a 1R resistor (with breakout cables and a breadboard) and connected my dmm in parallel to measure voltages (I thought the DMM was messing something up) and 1R was enough in series to make it work fine, it seemed. I tried then, experimenting, with a 0R1 and it stopped working again. I was able to measure (I don't remember in which way) around 180mA of peak startup current.

I then put everything back as it was (ie: no resistances, PB not working, PSU working), but I had been clumsy and accidentally disconnected the VCC cable from the VR module and… even with the powerbank everything was working!! I assume power trickled down from the TX-RX lines, but I don't get how or why I have this behaviour. I really want to learn something here.

Can you please give me any pointers? At this point I'm scared of VR modules.
Cheers

PS: I attached the schematic, sorry if it's partly Italian, everything should be there tho

[m k]

Is the Command part drawing as it should?

[Moriambar]

Is the Command part drawing as it should?

Hi, thanks for your reply.
I don't get your question, though...

[m k]

Is that part of the pdf as the real thing is actually connected?

[Moriambar]

Is that part of the pdf as the real thing is actually connected?

Yes, I'm sorry, perhaps it's drawn poorly. J7 is across C5 (which is there for debouncing) and it's connected to a switch (external).

[m k]

I saw different and wrong net examples around these voice modules.
So check that yours is correct.

Check also power bank voltage of non working situation.

[Moriambar]

I saw different and wrong net examples around these voice modules.
So check that yours is correct.

Check also power bank voltage of non working situation.

I think mine is correct. The purpose is to "sense" whether the module is active or not, thus doing different things in the mcu.
The problem is that with the PSU everything works fine every tiime.

I'll try and check the voltage from the bank whenever I get the chance (I have the circuit no longer), but why cutting the power line, hence getting power from the serial interface does work?
I suspect that's something to do with my cap, maybe?

[m k]

Something is clearly overloaded.
Powering from data line situation doesn't change to worse if you add extra power.

How is the VR connected exactly?
No pin names only, tell physical locations compared to other pins.