Voice recog without Alexa etc?

HavocXphere

Honorary Master
Joined
Oct 19, 2007
Messages
33,153
Reaction score
1,297
Location
Europe
I've got a fairly classic HA/lights setup. Keen to add some basic voice functionality - basic on/off 2 zone type deal so it can be pretty dumb. But without the usual Alexa & friends.

I've got a 6 mic array lying around that can in theory do it but my last attempt (Mycroft) at that was not amazing.

I don't mind hooking in something like GCP voice to text to translate it - so cloud in that sense is fine...just don't want the mainstream integrated solutions.

Anybody have any luck with something like this?
 
I used speechrecognition Python for a project but found anyone could turn on and off the lights. So personally I went with TensorFlow Python (Voice) since I could train my own voice and set the percentage to 98+. So only I could control the lights, curtains(beside if they imitate my voice).

There a TensorFlow.js also
Example: https://codepen.io/livecodestream/pen/JjGyoMv
Just set the probabilityThreshold to 0.9

I love to know what others are using, since I want to implement it with my new place, might be better and easier solutions.
 
Last edited:
I used speechrecognition Python for a project but found anyone could turn on and off the lights. So personally I went with TensorFlow Python (Voice) since I could train my own voice and set the percentage to 98+. So only I could control the lights, curtains(beside if they imitate my voice).

There a TensorFlow.js also
Example: https://codepen.io/livecodestream/pen/JjGyoMv
Just set the probabilityThreshold to 0.9

I love to know what others are using, since I want to implement it with my new place, might be better and easier solutions.
Amazing! So local is feasible. Thanks.

How did you do the hardware side? Rasp? Server? CPU? GPU TF?
 
Amazing! So local is feasible. Thanks.

How did you do the hardware side? Rasp? Server? CPU? GPU TF?
Raspberry Pi 3, was running smooth, no Issues. All it did was run my Gekko Miner, The voice(python) that posted to the RestAPI and then just read commands from the vps server, push it out through bluetooth(I see the wifi modules are cheaper now then the time I started).

But I ran the server on digital ocean($5 vps). Stores all the data and stuff. I used Arduino to make the boards and then 3d printed the mini cases for the switch, the curtain stepper motor.

1. Read voice command and send to server via RestAPI
2. Server processes Telegram messages and any voice request
3. It then send a notification(SignalR) to Raspberry PI to send the action to apply to the Bluetooth devices(was 5 devices)
4. The PI sends events to the Server to tell the statuses of the devices/relays.
5. The PI Sends Temperature and Humidity to the Server
6. The PI sends Water gauge pressure and Electrical feed status back to the server
7. The PI sends events of the motion sensors in each room
8. Server can send notification to raspberry PI to turn off geyser or lights.
9. Alerts are send to telegram bot, also u can ask telegram bot for statuses or even dim the lights.
/dim passage 50 to dim the passage lights by 50%, this was the most expensive module I bought.

With this setup my plan was always add more stuff to it, but never ever got time to expand on it due to time and cost. I just waiting for electrical certification from the seller and I am putting Sinoff Wifi Smart Switches and trying out the setup again. I keep falling asleep with the lights on, before I had a schedule that check if PC is on, my android phone status, time + motion sensor then turns off the light. Also winter is coming and going to be real lazy coming out from the warm blanket to turn off the lights! usually I have a bootstrap button, that calls rest api to turn off all lights and dim passage lights to 50% with a click of a button!
 
I wrapping up a fun project at the moment. Allowing a background movie play on my Microsoft Teams. So my camera -> OpenCV + TensorFlow or green screen -> fake camera loop -> Microsoft Teams

After that I starting the project from ground zero but using PI 3. I will share my journey.
 
Isn’t this built right into HA now? Shouldn’t need any third party?
 
I wrapping up a fun project at the moment. Allowing a background movie play on my Microsoft Teams. So my camera -> OpenCV + TensorFlow or green screen -> fake camera loop -> Microsoft Teams

After that I starting the project from ground zero but using PI 3. I will share my journey.
What do you aim to achieve? Basically looping video behind you during a Teams call?
 
Top
Sign up to the MyBroadband newsletter
X