Very new to scripting and coding, but taking notes.
A little about my machine… With the Raspberry prices going through the roof and almost non existent. I went the Dell Wyse 5060 TC route. I picked up 5 of them for like $50 shipped… I cracked one open and got to putting it into operation and my Linux/Ubuntu 22.04 Mycroft machine… I maxed out the ram with 16gb of ram and added a SATA 240GB SDSSDA 2.5 SSD from SanDisk and short 22 Pin SATA extension cable and lastly I added a new AX210 wifi 6E Card for M.2 NGFF Mini with two external antenna kit. They are basically a Raspberry pi6…
I had Mycroft working on it but I was having issues with that and ditched it for Rhasspy which also can run offline.
After 4 attempts I finely got rhasspy downloaded and running (I think) I can start rhasspy from the terminal with ‘rhasspy -p en’ and then access the localhost. And that is where I am…
My first thing is I am assuming (I know, bad to assume) that this runs side-by-side with Home Assistant or something along those lines as far as I can understand at this moment. But no info concerning, “you should install Home Assistant 1st or last” But almost all the documents and tutorials say that is works within Home Assistant. Like, Oh… And you don’t have to ditch Home Assistant.
The lines seem to blur when it comes to the offline part. I keep seeing ‘in the docker’, ‘it runs in the docker’, ‘inside the docker do this or that’. I did go to the doc site for the instructions to install Rhasspy, but chose the tutorials I followed which was actually easier to follow. The tutorial I followed advised that if you are only using the application on a single machine just for Rhasspy as a main server type application, Don’t follow the docker install. And my only use of my machine will be the hub/server to access and grow Rhasspy out to a really great voice assistant along the lines of Mycroft/Home Assistant.
I don’t know if I should have installed Home Assistant first.
If Home Assistant is open source like Rhasspy and can function offline as well.
What to do if I didn’t go the Docker install direction.
Can I even grow this project outside a docker.
Do I have to restart the project from a different direction.
I am basically at:
“So,…You managed to downloaded Rhasspy, Awesome! And it works to the point you can get to the Localhost, that is Double Awesome!!. And see all the cool UI Buttons, they are nice, right?. And you have a initial understanding of how a Voice Assistant functions, so you understand the TTS & STT and Intents and Intents handling, That’s a Start… But you have no idea what the next step is!”
So any guidance this awesome project community can provide would be great.
romkabouter has been, and continues to be one of the top resources on this board, so his recommendations always carry a lot of weight.
Here’s a good discussion of the different types of HA installs: https://old.reddit.com/r/homeassistant/comments/11dbsyk/docker_vs_haos_proscons/
Given that you’ve got Rhasspy running on your machine, I would suggest that you go the “Docker” route. Couple of reason, you say you’re new to linux, and seem like you’re interesting in learning more. Docker is really cool tech, and when I found out about it I ended up spending quite of bit of time playing with it. If also makes some things easier (and some things more difficult, sorry to say) with HA and Rhasspy. That being said, I’d still give it a go, set up Docker:
and then follow the instruction on the second half of this page:
If you go this route, you can leave your Rhasspy install as it is currently working.
Thanks for reading my post and tons more for the reply, and I will be sure to thank romkabouter as well.
In reference to docker… my gut said to stick with the docker path. but I didn’t understand it and I thought is was to be used when you are on a windows based machine and need to partition off a section to run like a Linux based program. And the path I went was very smooth aside from dependency errors. Once I got those ironed out. the 4 step DL was very smooth. But I will stick to DLing the docker as you suggested and then DL the HA OS as romkabouter suggested…
I’m sure I will be back should I run into another wall I can figure out.
I’d agree with the above. You want docker. It’s not just for windows. It’s the first thing I install on almost every Linux machine. It allows you to start containers inside your base OS. It’s super powerful. I’d suggest you watch a few YouTube videos on “what is docker”.
Prepare to enter the rabbit hole. Once I figured it out I was truly amazed by the applications of this software.
Honestly, if I was building your system with your specs from a fresh install of Ubuntu. I’d run HA in docker and (as stated above) Rhasspy as an add-on. If you wanted to run Rhasspy on it own you could run it in docker as well as a separate container.
I guess the confusion stems from the fact that Rhasspy doesn’t have any skills by itself, so it doesn’t know how to handle requests given to it.
The question is what do you want to do with it, A lot of people connect it to Home Assistant to allow Rhasspy control it, but that’s mainly a home automation software, and as you don’t already have it installed I wonder if it will actually do what you need (it’s not an mycroft like intelligent agent)
You can check this forum to see some examples of what people made with Rhasspy
$50 for all 5, not each. I think with the 2" sada cable, 16gb ram (2 x 8gb sticks), 240gb ssd, and wifi m.2 & duel antennas. I have $65 maybe $75 in the one i have running.
Being that the 5060 running ubuntu is now basically a pi4b +++. I think i did ok…
The will be my main machine for my HA & Rhasspy system to handle things in the house…
My thinking is I can run the others in a lesser config as they will be end points. Like the one i will put in my shop to control lights, music via an app, maybe my smart tv…
You may be right… I’m not sure as i have no experience with HA…All the other stuff i have played with was like Jarvis type projects where you have to code/script everything… and in those cases it was trying to follow what someone else did in a toturial or download some project from guthub. But they always seem to work maybe 3/4 as stated.
When I saw how checking boxes and adding small lines of intent (I think thats what it was), I thought that may be an easier way to go than trying to code/script major lines of code.
When i first saw Rhasspy, they made it sound like, “Add Rhasspy to anything and make your task/request much easier!”
Yes the documentation leaves the handling of the intents a bit of an afterthought, what rhasspy does is basically match your request to a template and make that template (or intent) available to programs to handle, only this handling is not yet that developed, it works with HA for home automation as it knows how to handle intents for home automation, but I don’t think it does assistant like things (also probably not the best system to build only an assistant as HA is made for home automation).
See this thread for someone who added Rhasspy to chatGPT to do kind of assistant things
This project supports a bit more assistant features out of the box: news, music (radio), timers, alarms, reminders, to-do and shopping lists, smart home (e.g. using open-source tools like openHAB), navigation, places, weather, Wikipedia, web-search, soccer-results (Bundesliga), a bit of small-talk
Didn’t try it myself (as it only does English and German), so I don’t know how easy it is to install or how well it works
HA is another rabbit hole, once I started playing around with it, I spent many an hour figuring out all of the cool things I wanted to have it do.
You do have an advantage with the “full” install of Rhasspy and using the Docker HA. I originally started with the Docker of Rhasspy, and one of the most frustrating parts of the set up with Rhasspy was the microphone. I had a “score” like you had with your hardware, I got some really good USB mics cheap, but because of how they were set up, and how sound works in Linux and docker, I lost many hours trying to get everything to work well (sampling rate wrong, low volume, garbled voice). When I switched to the “full” Rhasspy, I was able to set up the proper sampling rate and microphone levels, and now I can be down the hall in another room and get Rhasspy to respond to the wake work and recognize my intent.
As for “just add Rhasspy” I wouldn’t say that’s 100% true at this point, figuring out intents, events and automations will require some work, but if you’ve worked through getting rhasspy installed, I’m very certain that you’ll do very well!
I read you are still a bit on the “I do not know what I need or want” side, so I also want to suggest the Proxmox route.
Basically proxmox is a virtualisation toolkit, and you can add Virtual machines and Linux Container in it.
I have proxmox running on an Intel Nuc i5. Home Assistant is running in a VM on that, the Operation System version, there is a very easy script to create the VM and install HA.
You can then either use the Add On ore create another VM of LXC with docker and Rhasspy.
On my Nuc I also have a VM for Frigate, which is a NVR system and able to do object detections and stuff.
Whenever I want to try something new, I just create a VM or LXC and fiddle around with that. That way my core software is not touched.
Probably another rabithole, but I nice one so wanted to share it.
I would probably reserve one of the devices for just a Rhasspy server on bare metal.
It is a nice machine with AVX extensions, so it is possible to use DeepSpeech too, if that is off interest, which is not possible on a Rhasspy.
Proxmox will put a virtualization layer between the Rhasspy server and the hardware, so the hardware acceleration in the extensions might be lost.
That sounds great, but I’m leaning toward offline eventually. Then to only access the net for information when needed.
Before I moved toward Rhasspy I was playing with Mycroft and trying to integrate GPT3 for the online things and assistance with adding features to my project… But the guy that got it to working was major lacking with some details to make it work. So I dropped it!!
With proxmox it depends on the VM type as KVM will run a full VM and the emulation will cause overhead.
Pretty sure KVM emulates the full instruction set and AVX will be no problem its Arm that has a co-processor where the Neon is an accelerator, where cross arch takes a bigger hit.
LXC is a bit like docker and really just a fancy chroot using the current loaded kernel and hence usually near metal performance.
Dunno about the Dell Wyse 5060 AMD GX-424 but likely slower than a Ryzen 3 2200GE which I was hoping the Vega8 APU might also be able to be used but has been a nightmare with OpenCL and would seem a Vega10+ (Aka 11 onwards) really is needed, which is typical as that is only the next gen up.
The Ryzen 3 2200GE is a bit slower (30%) than the Rockchip RK3588(s) out there, which also gives Pi like energy efficiency and new are only approx £60+ P&P https://www.aliexpress.com/item/1005004941808071.html
I have been using the benchmarks for whisper.cpp and there is a good list of ML perf comparisons there.