Performance & Speed Setup

[exotic ledge]

Good evening all, I've messed arround with a few different setups and so far I have not really found a useable / speedy Voice setup.

Originally I had a Pi and a Nuc with a few M5 Atoms and performance was terrible.

Latest Setup was a dell optiplex as a server with a few M5 atom and same performance, maybe slightly quicker.

I've messed arround with the other voice satalites availible on pi hut and ali express but i've not really found anything that has both speedy response / a decent understanding of what i'm asking.

So i'm curious now to what everyone else has as their setup including on the software side.

[jaunty sage]

the voice processing speed doesnt really have much to do with the sattelite. they sattelite is not doing the processing. although a better one will likely produce a clearly sample for the processing to be fair.

what does your voice pipeline look like?

[exotic ledge]

Right now nothing, i'm currently in the process of doing a rebuild - so kinda just looking as best way to put it back in.

[haughty lynx]

Then what was it?

Atom Echo is practically the worse thing you can have for the satellite. Defined by proce. 🙂
What actually is bothering you?

[jaunty sage]

if you have the ram then using THIS over whisper may help. its pretty snappy and effective (for english anyway)

[haughty lynx]

Does it work good? Docker container image is broken right now, so i can't try it... 🙂

[jaunty sage]

Yup I have been using it as add-on on n150 mini PC and its been smooth

llm is running on another system but STT and TTS are running on the mini pc

even with longer message its still pretty sharpish

i never got it working cos of some misc issues (i think with my setup) but you can gpu accelerate the normal container version which apprantly makes it insanly quick

was running whisper large-v3-turbo on a 5060ti (on the llm running machine) previously for stt and its probably better tbh. and frees up some vram on the other system 🙂