Speech-to-text for Home Automation without Google/Alexa

Dear all,

I would like to use my MATRIX Voice for my home automation solution with openHAB.

I would like to avoid Google and/or Alexa (AVS) and rather use an open source Speech-To-Text system (internet-based would be okay).

The text should be sent via MQTT or other protocol to my openHAB for further processing.

I’d be happy about hints, tips and tricks!

Will try to keep this post updated with what I am working on.

1 Like

Hi,

maybe you can use the malos-wakeword service, it is a service using pocketsphinx and is runing totaly on the raspberry.

https://github.com/matrix-io/matrix-malos-wakeword.

I tried this last weekend on my creator and it worked pretty well.

as described you can easily create own English models:

i have used Snowboy previously on RPi and it’s great. would work with Voice as well.
https://snowboy.kitt.ai/

Hmm can snowboy detect different sentences or only single wakewords? My tests only worked with single wakewords e.g. Alexa / Jarvis, etc.

Well, the way I did it was to use it for Wake word recognition. Then, I did two different things as trials. One was to pass a second phrase to snowboy for further processing. The second was to use Wolfram online to process further speech.

sounds interesting. have to check wolfram online :slight_smile:

Can look at snips to do the same thing using the matrix voice as the microphone.

@tschmidty Do you get it managed to run snips with matrix voice on a RPi?! I´m stuck setting up the microphone correctly.

It works well if you set up the audio server like this

snips-audio-server --nomike --hijack 0.0.0.0:1234 & arecord -r 16000 -c 1 -f S16_LE | nc localhost 1234 &

Apparently, snips can’t use the mic directly. But follow the basic matrix instructions, install snips and run that command and you should be set.