I've noticed that when doing text file to speech that the dsnote process will actually use all 12 of my cores and it will hold them hostage for quite some time.
I load up a text file of 22K, using "English (Mimic3 LJ Speech low) / en" as it seems to be somewhat light weight while still giving a good sound. Almost no cpu is being used before I start. When I press Read, then all 12 cores show activity (which can be seen using xosview). "top" shows my load to be 8-9 and dsnote is using 730% to 850% of cpu ... and it uses it constantly! I can hit "pause" and it still using all of the cpu.
Why? Is there anything that can be done about this?
To add to the mystery, on the 22K file I noticed that the processing stopped after about 1m50s. After that I could pause, play, reposition and play again and no significant cpu was used, leading me to believe that the program was probably doing a background conversion for text to speech into an internal buffer (memory usage did go up some).
Is that guess correct? FWIW, I am using the hardware acceleration (Nvidia CUDA) if it matters.
Oh, under Advanced settings I see "Number of simultaneous threads" and it's set to "auto". Changing that setting to 4. It wants a restart, so closing the app. "dsnote" is not running according to "ps -ef". Restart.
It still has my 22K file in the "notepad" on restart, hit Read ... and sigh, it's still using all 12 cores at over 700% cpu. The heavy CPU usage lasted 1m48s, so the same. Maybe that setting is buffered per file?
Clearing the notepad buffer. Restarting the app again. Loading a new 6K text file. Read. Sigh, it's still using all 12 cores and the setting still shows 4.
So the real bug is the app doesn't use the "Number of simultaneous threads" setting correctly.
It shouldn't matter, but possibly useful system values:
cpu: AMD Ryzen 5 2600X (12 cores)
video: NVIDIA GeForce GTX 1660 Ti (nvidia driver v580)
os: Ubuntu 24.04.4
dsnote: v4.8.3 (in flatpak)
I've noticed that when doing text file to speech that the dsnote process will actually use all 12 of my cores and it will hold them hostage for quite some time.
I load up a text file of 22K, using "English (Mimic3 LJ Speech low) / en" as it seems to be somewhat light weight while still giving a good sound. Almost no cpu is being used before I start. When I press Read, then all 12 cores show activity (which can be seen using xosview). "top" shows my load to be 8-9 and dsnote is using 730% to 850% of cpu ... and it uses it constantly! I can hit "pause" and it still using all of the cpu.
Why? Is there anything that can be done about this?
To add to the mystery, on the 22K file I noticed that the processing stopped after about 1m50s. After that I could pause, play, reposition and play again and no significant cpu was used, leading me to believe that the program was probably doing a background conversion for text to speech into an internal buffer (memory usage did go up some).
Is that guess correct? FWIW, I am using the hardware acceleration (Nvidia CUDA) if it matters.
Oh, under Advanced settings I see "Number of simultaneous threads" and it's set to "auto". Changing that setting to 4. It wants a restart, so closing the app. "dsnote" is not running according to "ps -ef". Restart.
It still has my 22K file in the "notepad" on restart, hit Read ... and sigh, it's still using all 12 cores at over 700% cpu. The heavy CPU usage lasted 1m48s, so the same. Maybe that setting is buffered per file?
Clearing the notepad buffer. Restarting the app again. Loading a new 6K text file. Read. Sigh, it's still using all 12 cores and the setting still shows 4.
So the real bug is the app doesn't use the "Number of simultaneous threads" setting correctly.
It shouldn't matter, but possibly useful system values:
cpu: AMD Ryzen 5 2600X (12 cores)
video: NVIDIA GeForce GTX 1660 Ti (nvidia driver v580)
os: Ubuntu 24.04.4
dsnote: v4.8.3 (in flatpak)