Are there any advices how to optimise neon OS?
are there any options to outsource operation to other devices, like a central server?
I have a mark II but I don’t experience a very responsive conversation.
Hoping for options to make conversation with neon more fluent
Yep! First, set up zram. It helps a ton with the Pi 4 2GB inside. There’s a post here in the forum somewhere explaining the steps.
Next, make sure you’re on a USB 3.1 SSD. Then, set up your STT:
I don’t have a post like this for TTS yet, but you can give this a try to start in neon.yaml:
Finally, make sure you’re not loading anything with Vosk by default or explicitly. Vosk is a memory intensive bit of software and we don’t have much memory to spare. Look at wakewords and STT fallbacks.
If your offloaded TTS/STT are down or unreachable you will see a ton of latency as the local fallbacks are used. With the above changes, though, you should notice a much improved conversational experience.
Hope this helps, and happy to elaborate!
what shall I use instead?
Do you refer to
This means that you probably want to try to set an on-device plugin such as Vosk or Silero as your fallback. That way, you can still use your Voice Assistant in case your server is down or unreachable for some reason. Just keep in mind that an on-device STT option is usually lower quality and consumes a great deal of resources, so make sure your Voice Assistant device can handle it.
I haven’t tried it personally, but the OVOS folks have recommended Silero as a fallback over Vosk because of the memory consumption.
Silero, Vosk STT and Vosk WW are three different things.
Silero is a
VAD plugin, where as Vosk can do STT and WW detection.
Silero works better than webrtcvad, but that is for detecting when you are actually talking.
Vosk does take a lot of resources though