Best examples for offline stand-alone ESP32 voice interface

Hi there!

Are there any good ressoruces for a stand-alone, offline voice interface just with the matrix-ESP32 hardware?

In the past I used the SNIPS algo but what is the way to go today, after SNIPS was accuired by Sonos? Are there any open source projects for ofline ESP32 with Matrix?

This one is jut a setup tutorial but not a voice example.
https://matrix-io.github.io/matrix-documentation/matrix-voice/esp32/

Many Thanks.

I’ve had my Matrix for over 2 years and I have literally never used it before. The installation process seems unnecessarily complicated. I actually made attempt last night to get this working to no avail. I’m not trying to be negative no, but I think I may have to cut my losses.

Check Rhasspy:
https://rhasspy.readthedocs.io/en/latest/

Community:

ESP32 Voice, check my repo. It acts as a MQTT mic and supports Rhasspy:

I can PM you my address if you want to get rid of it :wink:

I may take you up on that.

Also, appreciate your URL. I think what some people at Matrix dont understand is that people who were expecting a user-friendly device dont know this kind of stuff. I dont know PlatformIO nor do I know MQTT. This is all quite frustrating. However, in the end its my fault for making assumptions.

I’ll let you know if I plan on selling it.

I partly understand your frustration, I had some startup troubles as well.
But the device itself is clearly one to be use by developers, tinkerers and all other sorts of people with an affinity for these kind of devices.

If you do not have any background with such device, I can only give you one advice:
Never buy a device when it’s not in a casing. If you see any kind of chips on the outside of the device, do not buy it.

I used pocketsphinx in one of my projects. It decode words or sentences, where you can stablish your callback using c++.

Tell me if you want more info

Hi Romkabouter,
Nice to hear that you are happy with the Matrix voice.
I sorted out my Matrix Voice about a year ago because most of the features didn’t work.
As a hardware specialist, have you now got the functions of Matrix running?
Does the audio output now work with values other than 44100Hz / 16 bit?
How did you get the acoustic echo cancellation up and running?
Beamforming, acoustic source isolation, noise suppression now also works now?
I’m curious how it works.

Not at all, I resample incoming audio to play at 44100 16 bit (only supporting 16bit)
I do nothing with the other stuff you are asking about :slight_smile:
This is becasue of my project is for the ESP32 and I think those advanced functions are too heavy for that chip and most likely will rely on a Pi attached.

I do have wakenet running, wakeword detection on device:

Ok, that’s a pity that you don’t use all the great voice features, AEC and noice cancellation and these things are the features that make the MatrixVoice stand out. Do you know anybody who made these features work? After 3 years of development and community, you’d think that these features would be up and running.

MatrixVoice is just another micarray with leds and I think ReSpeaker has the advantage over it.
Also more can be found on those topics that feature the ReSpeaker, so I do not think there are a lot of developers using the Matrix Voice. And certainly not the ones with knowledge about AEC and such.

The esp32 chip is what makes this device different in my opinion.