Piper GPU Acceleration

apophis · 03-23-2025, 08:54 PM

I love the new pre-computed TTS feature that came in 7.21, but with piper it is far too slow to convert even just the good books of my fairly large library.

Per the Piper documentation, Piper supports CUDA GPU acceleration but it doesn't seem to work with Calibre.

Is there any way I can enable this with Calibre?

I am willing to go so far as to write my own plugin if that is necessary.

Thanks.

kovidgoyal · 03-23-2025, 10:24 PM

piper in calibre is a standalone executable, you could build your own version that supports GPU acceleration and replace it in the calibre install. IIRC the GPU acceleration comes from onnxruntime not piper itself, so you would need to build that against cuda then build piper against that, or just replace the onnx dlls in the calibre installation folder might be sufficient, not sure.

apophis · 03-26-2025, 03:17 PM

Hello,

Unfortunately, despite what the piper documentation suggests, CUDA gpu support is not in the latest release version (2023.11.14-2)(or whatever prior version Calibre uses). Thus replacing the onnxruntime dlls is not sufficient in of itself. (piper.exe --help will include "--use-cuda" if it is supported)

More over, even after building the latest version from source with (I believe) CUDA support enabled, the executable crashes silently when loading the voice data. It works fine w.o CUDA, so I don't know if I built it wrong or it is just unusable.

Regardless, several github issues on piper suggest it would be slower than CPU anyway. I don't know if this would hold true for very large data like ebooks, but I can't get it working anyway.

Thanks for you help and quick response.

03-23-2025, 08:54 PM	#1
apophis Junior Member Posts: 2 Karma: 10 Join Date: Mar 2025 Device: none	Piper GPU Acceleration I love the new pre-computed TTS feature that came in 7.21, but with piper it is far too slow to convert even just the good books of my fairly large library. Per the Piper documentation, Piper supports CUDA GPU acceleration but it doesn't seem to work with Calibre. Is there any way I can enable this with Calibre? I am willing to go so far as to write my own plugin if that is necessary. Thanks.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
GPU Use when converting book format	NugentS	Conversion	7	03-31-2023 01:42 PM
GPU Temperature showing up in ebook viewer	misterioso	Viewer	4	07-19-2020 06:40 PM
T68 - Hardware graphics acceleration	emmen	Onyx Boox	6	12-29-2014 04:39 AM
Visture V4 RK3066 Quadcore GPU	Dvalin21	Android Devices	0	09-03-2012 09:39 PM
Android Honeycomb Port For Nook Color Gets Graphics Acceleration	Nathan Campos	News	0	01-30-2011 01:11 PM

03-23-2025, 10:24 PM	#2
kovidgoyal creator of calibre Posts: 45,345 Karma: 27182818 Join Date: Oct 2006 Location: Mumbai, India Device: Various	piper in calibre is a standalone executable, you could build your own version that supports GPU acceleration and replace it in the calibre install. IIRC the GPU acceleration comes from onnxruntime not piper itself, so you would need to build that against cuda then build piper against that, or just replace the onnx dlls in the calibre installation folder might be sufficient, not sure.

03-26-2025, 03:17 PM	#3
apophis Junior Member Posts: 2 Karma: 10 Join Date: Mar 2025 Device: none	Hello, Unfortunately, despite what the piper documentation suggests, CUDA gpu support is not in the latest release version (2023.11.14-2)(or whatever prior version Calibre uses). Thus replacing the onnxruntime dlls is not sufficient in of itself. (piper.exe --help will include "--use-cuda" if it is supported) More over, even after building the latest version from source with (I believe) CUDA support enabled, the executable crashes silently when loading the voice data. It works fine w.o CUDA, so I don't know if I built it wrong or it is just unusable. Regardless, several github issues on piper suggest it would be slower than CPU anyway. I don't know if this would hold true for very large data like ebooks, but I can't get it working anyway. Thanks for you help and quick response.

Advert