I like to share my code for a voice assistant, it’s been hard to actually make it work, I modify a Bluetooth speaker and add it inside for more appeal case.
I had two versions, one for
esp32 wroom without media player, timer, and continuous conversation, lights, without micro wake word
ESP32 dev
Max98357
Inmp441
led strip of 4 segments
and an esp32-S3
with media player, vol up and down, timer, continuos conversation, lights, mute without micro wake word, sound on vol up and down, mute and timers, also sound on boot
ESP32-S3 N16R8
Max98357
Inmp441
2 buttons to control volume of media player
led strip of 4 segments
You are using many outdated solutions.
If your board has PSRAM, it makes sense to create two audio pipelines, so music will always play in parallel and be muted during announcements.
Additionally, there’s no point in disabling the wake word; it would greatly simplify the operational logic (see the code for VPE or ReSpeaker Lite).
I agree but also depend on the hardware you are using ReSpeaker use different hardware for audio output, and 2 microphones just like voice PE
in this configuration you are only using a Max98357 and a Inmp441
It’s not critical, you can take a look at my solution. I’ve been testing it for a week and haven’t noticed any issues. However, I removed everything related to the addressable LED since I’m not a fan of it.
The files are on github, so you can use links to avoid downloading them manually. Or replace them with your own.
I saw a post about led, you can return the original led control script and running it in triggers without problems. Don’t forget to remove my indicator and related actions.
really great! The fact that it stayed listening shortly after the action is really convenient and functional. thanks to everyone!
Only problem… once the timer rings I can’t turn it off with my voice.
I don’t know if this is due to the code change for the LEDs and buttons or if it doesn’t work by default.
In my configuration it definitely works (using ok_nabu). Theoretically, adding new actions should not affect the operation. If it doesn’t work, try removing the LED triggering for the timer_ringing switch, this will free up some resources.
If you are talking about the word “stop” then it should be added additionally for timer_ringing triggers. Use the commands micro_wake_word.enable_model: stop and micro_wake_word.disable_model: stop
I can also recommend expanding the word list for the HassNevermind intent:
This will explicitly stop the pipeline at the listening stage, instead of waiting for time to expire. This often sounds pretty natural, especially when interacting with llm.
Here is my example, do similar for your language
