Year of the Voice - Chapter 4: Wake words

Blog
62

@greggotcher there was a bug in the example Google Colab notebook that was fixed last night. If you try again with the updated version it should work correctly.

2 Likes
63

Same with Whisper

64

I saw that in the stream. Where is the Assist Microphone add-on located?

65

It’s in the first post.

66

Hello,

First of all, I am impress by the progression of this year of the voice.

I have a question about the openWakeWord. As it’s seems to use Tensor flow ligth, did you plan to support the Coral USB Device for finding the wake work in the stream ?

Thanks,

1 Like
67

I am all ears :wink:

68

From my understanding of the Coral, this will not be possible unfortunately. Some of the TFLite operators used by the Google speech embedding model (what openWakeWord is based on) are not supported by the Coral. For example: FusedBatchNormV3 and LeakyRelu. Perhaps someone with more ML knowledge would know how to work around this.

4 Likes
69

Thanks for the effort you guys put into the Voice-Assistant :slight_smile:
Now - my ESP32 S3 Box can finally be what I had planned for it to be.

Unfortunately, it seems that there’s still a conflict when using the Microphone / Speakers in combination with the Display… but I am pretty sure, that this is also just a question of time when this might be solved… :slight_smile:

I have also plenty of other ESP projects in the pipeline, and I am looking forward to turn them not only into Bluetooth-Proxys and devices that controll a couple of LEDs… but similar to the StarWars Droid, they should then also act as Voice-Controller :slight_smile:

2 Likes
70

Is there currently support (or plans for supporting) Nvidia CUDA with GPUs?

71

Given Assist can be configured as the default assistant on Android devices (chapter 3), is there any way to link this wake word functionality to Android? i.e. trigger “Assist” on an Android device using the wake word?

72

Not within Home Assistant, but I imagine something like what has been done with the Whisper Docker container could be done for openWakeWord: Home Assistant - Enabling CUDA GPU support for Wyoming Whisper Docker container | tarball.ca

2 Likes
73

Can we have multiple wake words to work with our mulitple voice assistants? I currently have one assistant configured for home assistant activity, and one goes to chatGTP. It would be great to be able to give these different wake words so automation goes to Jarvis, whilst general questions goes to Deep Mind - or whatever…

3 Likes
74

interested to hear if you figured out where to map things to. I have openwakeword and homeassistant in different containers but cant seem to figure out where to drop the custom .tfilte file. HA sees openwakeword cause i can select from the pre-loaded 5 but cant seem to find my custom one.

75

Yes!
I needed to map a volume to the /custom folder on the docker container.
This is in my second volume mount below.
I also added the --preload-model 'hey_kelsey so that my wake word was the active one.
I also added the --custom-model-dir /custom so that the volume map worked.
I saw this in a random post on discord and it got me in the right direction.
I also made sure to copy my .tflite for my custom wake word into my mapped volume.
Here is my current stack.

# Open Wake Word Dection 
  openwakeword:
    container_name: openwakeword
    image: rhasspy/wyoming-openwakeword
    command: 
      --preload-model 'hey_kelsey'
      --custom-model-dir /custom
      --model 'alexa'
      --model 'hey_jarvis' 
      --model 'hey_mycroft' 
      --model 'alexa'
      --model 'hey_kelsey'
      --uri 'tcp://0.0.0.0:10400'
    volumes:
      - /home/pcwii/docker/openwakeword-data:/data
      - /home/pcwii/docker/openwakeword-data:/custom
    environment:
      - TZ=America/Toronto
2 Likes
76

Can someone please help me with wiring and config?
I have all parts and followed instruction but i cannot get it to work
I’m using INMP441 and esp32 for sound input but home assistant doesn’t see any sound input

image
image638×1020 43.6 KB

Wiring:
SCK > GPIO19
SD > GPIO23
WS > GPIO33
L/R and ground > Ground
VOD > 3.3V

Using following config:

Is this the way or i do something wrong?

77

worked like a charm.

1 Like
78

(Update: think I found my answer with “Assistant Microphone” from the add-on store, but leaving the rest of the question up for any other advice)

Hope this isn’t already answered, I’m pretty new here.

Is there a way that I can plug in a usb microphone to the Raspberry Pi 4b that is running Hass.io to listen for wake words and speech? I’m having trouble figuring out if this is feasible or not.

I’m trying to create a “Magic Mirror” from Shrek and would love to utilize the custom wake word detection of “Mirror Mirror on the wall”. So the ideal plan (considering a ~2week deadline) is to plug in the mic I have: (this one) to the Pi and use it as a speech detection microphone that then triggers an automation to cast local media files to a chromecast in response. (I have custom .mp4s that I want to play for each defined trigger sentence)

Any help understanding my options for microphone input to the main Pi running this stuff would be greatly appreciated. I’d be open to alternatives like using an android phone, ipad, Mac or PC as an alternative mic input as well. Thanks!

80

Anybody gotten this working with a google voice hat yet?

I have one somewhere in the house. Was using it for logitech media server client. I wonder if I could leave that intact and use voice also.

81

Anyone else having this issue compiling the ESP box code from here?

82

Could we also use a M5Core2 as a voice satelite with ESPHOME?
If yes, could we get the display working at the same time?