I run it on a raspberry pi 3B+ with a SEEED 2 mic hat. I am very impressed and recommend it to anyone with an interest in offline voice.
I have also tried the server/satellite approach but found that standalone devices worked better for my purposes.
Yes - I have not tested it out yet but the author of this project suggests Llama derivatives like Vicuna. I am excited to see how this project evolves alongside Homeasisstant’s voice goals. The author of Rhasspy is working for Nabu Casa so im sure that will grow too!
Unfortunately Mycroft has been discontinued by its management team and is heading towards deprecation. This project is starting up now and has a strong initial release that works well. Mycroft was intended to be used with a cloud backend hosted by the Mycroft team (that being said they did eventually open source their backend but it was not intended for use with single instances).
Willow is designed to work with very low power/cost hardware (esp32-s3-box) and either homeassistant or openHAB right out of the box.
It is very functional especially for DIY-inclined people! The creator has been hired by Nabu Casa (the creators of Homeassistant) to implement voice services for their Year of Voice.
This iteration of Rhasspy (2.5) is quite mature with good features. The only caveat is that it cannot handle wildcard intents. It can have multiple slots for a sentence (e.g Turn on the (kitchen|bedroom) light), but not a wildcard (e.g. Turn on the * light).
Rhasspy 3 has been released as a development build but is still quite early in its lifecycle.
I have never heard of this! Thanks for sharing