Year of the Voice - Chapter 4: Wake words

@sparkydave
It looks like you’ve created the share subfolder in config directory
You need to use the share folder which is at the same level as config directory

I tried to create a fork and upload my file, but I stack on the second step. Maybe some one will add this for our community.
Link to Google drive: Hey_Morgan.tflite - Google Drive
Thank you
My Rum Voice assist bottle almost done, just need to fill up with acrylic water ( spice rum color) and that’s it .




8 Likes

Is there way to add Home Assistant Cloud TTS and STT to Wyoming Protocol?

Anyone worked out how change the default “I dont understand” message?

Just a FYI for anyone using the vscode addon… It doesnt like yaml config for response files in custom_sentences\language.

For anyone else, here is the response from discord on this. I have not tried it.

create a file called custom_sentences/en/errors.yaml and add this to it

language: "en"
responses:
  errors:
    no_intent: "Sorry, I don't speak Pleb"
3 Likes

Thought I would add this to the chat as well.
I know there are many individuals who have attempted to use a wakeword with the Atom M5Stack and have run into issues where after a time it stops responding and it will only respond after you toggle the “use wake word” switch within home assistant. Apparently this is know and there is a thread on github that discuss this. One thing I found on github is a bit of code that I attempted and I have to admit that my M5Stack has been much better since I implimented this code. All I did was past the code at the bottom of my esphome configuration for the m5Stack then restarted HA and the unit has been running and responding for almost 24 hours without me needing to do anything. If anyone is interested in attempting to add this before there is an offical fix here is the code.

interval:
  - interval: 1s
    then:
      - if:
          condition:
            api.connected:
          then:
            - if:
                condition:
                  and:
                    - switch.is_on: use_wake_word
                    - not:
                      - voice_assistant.is_running
                then:
                  - voice_assistant.start_continuous:
7 Likes

What’s the ESP32 board you are using?

It’s a esp-wroom32 mini
Same like the regular, but less GPIO’s
Got them from Aliexpress long time ago. It was a deal 10 for $9. That was the last one from that order :grin:

1 Like

I like these as super small esp32-s3 as in the esp32-s3-box super small format and great price.

1 Like

That did the trick! Thank you.

1 Like

I add a single LED to my esp32 board using GPIO23 and 220 Ohm resistor for visual when Voice assist is ON, but can’t program this light, for some reason t’s does not recognize my code. I use simple code from ESPHome site.

light:
  - platform: binary
    name: "Voice Assist On"
    output: light_output

output:
  - id: light_output
    platform: gpio
    pin: GPIO23

What I am doing wrong?
TY

Is it possible to use “microphone” and “speaker” from different devices for wakeword activated assistant?
Has anyone managed?

2 Likes

I would like to use the microphone and speaker and possibly the volume control/leds from old alexa devices with an esp32. Does anyone have circuit diagrams or have seen these done before?

Hello, is there a section here in the community just for Voice Assistant? That would help a lot now.

Another question, my Intel Core i5-4250U is struggling to work with the Whisper add-on, when I use the Raspberry satellite the CPU usage goes to 100% and returns nothing, does nothing. Is my CPU too weak to do this? Is anyone having this problem?

3 Likes

Hi, thanks for sharing this, have added it to my config.

Do you know if it’s possible to add some kind of service call to this? In others words, when Assist responds with the error ‘Sorry, I don’t understand pleb’, can I get HA to do something else, like TTS, or turn on a toggle etc?

I’m asking because I’m currently using satellite with just a mic, and am trying to use my Google nests via TTS to give a response, but so far it only works with custom sentences. So I’m trying to find an alternative way to trigger the TTS.

I’ve tried, but not got it to work so far.

I have not tried. I saw the error answer on the voice assistant discord for HA. It may be good to ask the question there.
I have done automation responses with custom intents but haven’t tried them with the error. If I get a chance I will try that.

1 Like

Thanks very much, would greatly appreciate it if you could figure out a way to make a service call with the error, or if you have another way that an automation can be triggered or similar when the no_intent error is called :slight_smile:

Likely it would be better for the community to create custom commands than wake words.
Because adding a head model is superlight you could have a ton of commands that just trigger mapped mqtt.
Simple brute force, but because its no so much hassle to record a custom verifer, you can and likely would be quite accurate on very modest hardware.
No ASR needed as each command just triggers a script or REST Api or whatever.

A KW must proceed a command in whatever timeout windows and you can only submit with accompanying benchmark results.
The head files are very small and over time the community could build up quite a library and if a command does not exist just create one.

I received my esp32-s3-box3 through Ali (can someone explain why they chose baby blue for the docking station :thinking:). The device works out of the box - including an option to teach ir codes to automate airco based on the built-in humidity and temperature sensors.

Goal is however to use this as a HA voice assistant. Did anyone get this to work in esphome? Willing to share yaml?

Tried to play around with the regular s3-box examples but that’s so far not working.