Specifically, I have a piper service running on an old Mac mini so I can have a GLaDOS voice, and it is painfully slow. I only use it occasionally for fun with my kids, but even then I typically run the first play through a different speaker so it can cache and not take forever to play for them. Is it possible to offload the processing work to the speaker for a faster performance?
#Will the Voice PE be the one running voices?
1 messages · Page 1 of 1 (latest)
The speakers are running microcontroller processors, they don't have much horse power to process tts. I think the only way to get faster tts would be a more powerful processor. My piper tts is near instant, but it is running on a pretty powerful processor 🙂
That makes sense. Upgrading stays backlogged for now then. Thanks!
why not try a cloud service for that purpose?
Because I don't want to pay for any part of my setup (aside from devices) unless it's absolutely necessary
You can setup a local ddns with a ping uptime script. So you can use a other none 24/7 pc as faster system when its on.
STT/TTS are pretty computationally intensive. you kinda have to pay for them one way or the other: with subscription/usage fees, with hardware or with time
Im on my cloud trail and going to keep it. Cloud ai is so much better in picking up the right sentence and voice quality. You can make automations it fails back to onprem system if inet is down.
I don't mind paying for better hardware, but that's a low priority thing since it's not critical to my system. Eventually I'll upgrade.
That sounds more complex than I think I am willing to start working on. Plus the PC that's running the voice right now doesn't typically do much else. A couple of other containers that are mostly just scripts.