It would be awesome if we could have multiple voice assistants in operation simultaneously. So if I say “ok nabu” I get a nabu casa pipeline for controlling the house and if I say “OK Jarvis” I get a ChatGPT pipeline. Others have described how it would be useful to have different languages for bilingual households
This actually briefly almost worked by mistake Wyoming ignores second assistant's wake word. · Issue #101942 · home-assistant/core · GitHub (at least the open wake word side did).
Provided the hardware is capable of running multiple wake-word models it would seem trivial to implement. Just mandate that each active pipeline have a different wake-word and once a wake-word is detected forward the speech to the appropriate pipeline/wyoming endpoint.
I’d love to see this. I actually have bought multiple of those Atom Echo devices and plugged them into USB outlets in most of my rooms, and they work decently enough as generic house control.
Recognizing different wake words on the same hardware, would allow a different pipeline to be associated with it, and to be route to something like my Ollama instance for general LLM interaction.
I think this is useful for safety and not only for “fun”: I have three mini-me running around my apartments and the oldest is already capable to ask google to start music, light etc.
I do have also electric shutters in my house and I think this should be used only by adult (It’s not fun to be locked outside). Having multiple wakeworks can be a good solution.
For example: “Hey Jarvis, lower the shutters” for me and my wife. But this should not be available for my daughters wakeword (Hey Casita - for example using Encanto disney movie).
I think this a great use case - but not for multiple wakewords … because it won’t take your kids long to work out that they can also give Jarvis commands.
Instead this seems ideal for Voice Assist to recognise the speaker.
I’d like to suggest that support for multiple wakewords could be used for more than just running multiple pipelines.
Commercial voice assistant smart speakers like Google Home or Alexa can cancel/stop alarms, or stop playing music, when the user says “Stop”. Home Assistant’s voice assistant should also have this capability.
What if we were to support multiple wakewords, and use “Stop” as a secondary wakeword that could stop timers, and perhaps send commands to HA to stop playback on media devices?
with some help from Gemini I was able to finally do it.
step 1: alsa config to split input device. in my case it is Poly CL-3200M speakerphone.
# /etc/asound.conf
# Define the shared microphone input using dsnoop
pcm.mic_shared {
type dsnoop # Use the dsnoop plugin for sharing capture
ipc_key 1024 # Unique IPC key for this dsnoop instance
# (must be different from any dmix ipc_key)
ipc_key_add_uid false # Set to false if accessed by different users/root
# Set to true if only accessed by the same user
# For Docker with host devices, false might be safer initially
slave {
pcm "hw:P3200M,0" # The actual hardware device from arecord -l
channels 1 # We want mono input for voice
rate 16000 # Your desired sample rate (e.g., 16kHz for voice)
format "S16_LE" # 16-bit signed little-endian (common for voice)
period_size 1024 # ALSA period size in frames
buffer_size 8192 # ALSA buffer size in frames (e.g., 4-8 periods)
# Larger buffers can help prevent xruns if RT is an issue
# but increase latency.
}
# Optional: If your P3200M is stereo but you only want the left channel
# (assuming channel 0 is left, channel 1 is right for the hardware)
# If P3200M is mono, this bindings section is not strictly needed
# but doesn't hurt. If it's stereo and you want to pick one channel,
# make sure 'channels 1' is above in the slave {} block.
# bindings {
# 0 0 # Bind channel 0 of this mic_shared device to channel 0 of the slave (hw:P3200M,0)
# }
}
# Optional: If you also want to make this shared mic the default ALSA capture device
# (Be cautious if other applications expect a different default)
# pcm.default {
# type asym
# playback.pcm "plughw:P3200M,0" # Or your preferred default playback
# capture.pcm "mic_shared"
# }
# ctl.default {
# type hw
# card "P3200M"
# }
# If you want a simple default playback to your P3200M speaker for testing
# (used by your Docker 'aplay' command later if you change -D)
pcm.!default {
type plug
slave.pcm "hw:P3200M,0" # This assumes P3200M is also your desired speaker
# If not, use the correct playback device here
}
# This ensures control elements also default to P3200M
ctl.!default {
type hw
card "P3200M"
}
Now I can run one open-wakeword and as many wyoming-satellites as I like.
step 2: patch wyoming-satellite to also install alsa-utils:
