search

rhasspy / piper

A fast, local neural text to speech system
https://rhasspy.github.io/piper-samples/
MIT License
6.44k stars 473 forks source link

How fast is piper? #235

Closed francqz31 closed 1 year ago

francqz31 commented 1 year ago

Hello Authors, I wanted to ask can piper run in real time ? how fast is it on an rtx3090 or any high end GPU? I'm asking since I will use piper in a ai customer service experiment so it needs to be ran in human real time, if you know what I mean.

Thanks in advance!

StoryHack commented 1 year ago

How fast it can generate depends on the quiality of voice (low vs medium vs high) and the cpu, but yes. Even high quality voices generate much faster than realtime on my moderate system. I've never used the pip-installable version so that I could run inference via GPU, but that should be much faster than cpu.

synesthesiam commented 1 year ago

I have some benchmarks here: https://github.com/rhasspy/piper/issues/33

A lower "real-time factor" (RTF) is better, so even on a Raspberry Pi 4 you can get faster than realtime with low/medium quality models. On my AMD 5950X processor, even the high quality model runs over 12x faster than realtime (RTF = 0.082).

francqz31 commented 1 year ago

Thank you so much for both of your answers!