Calibre 8.9: TTS output, and where are TTS files stored?

Moonbase59 · Yesterday, 04:44 AM

So I’m using Calibre 8.9 on EndeavourOS/Cinnamon (rolling Arch-based distro), and I’m experiencing some oddities with the viewer’s TTS output.

I have a working speech-dispatcher output, using a selected set of Piper voices (English and German), and things like spd-say »This is a test.«, or Foliate work nicely.

Now in Calibre’s ebook viewer, when selecting »Read Aloud«, it immediately started downloading a voice called libritts (United States) [High Quality]. I can also download other Piper voices.

Click image for larger version

Name: Configure Read aloud_002.png
Views: 14
Size: 171.0 KB
ID: 217742

Where are these stored? I don’t really want to download & store all voices I use twice.

When using the suggested The Piper Neural Engine, a reading progress is shown in the book (blue highlight) but I get no sound, whatever output device I select.

Click image for larger version

Name: Childhood's End [EPUB] — E-book viewer_001.png
Views: 6
Size: 250.6 KB
ID: 217744

When switching to my already-installed The Speech Dispatcher Engine, I get a list of my installed voices, can select one, and it reads the book text in the selected voice—but without any indication of position, and no page turning.

Click image for larger version

Name: Configure Read aloud_003.png
Views: 10
Size: 163.6 KB
ID: 217745

Would be nice if this can be solved, preferably with the reading progress shown.

kovidgoyal · Yesterday, 07:37 AM

If you want reading progress you have to use the piper engine. speech dispatcher doesnt have any. It's output goes via Qt Multimedia which in turn goes via ffmpeg which goes via pulse audio (or pipewire) which goes via the linux kernel to your hardware. something in that chain is broken in your system. Fix it and you will be fine.

Moonbase59 · Yesterday, 10:19 AM

Yeah, I know, unfortunately that can’t work with SPD as it is.

Thanks for the short insight to the pipeline. I’m using Pipewire, so I installed qpwgraph to check what happens, and surely enough, it had selected an output called "System default (currently Comet Lake PCH-LP cAVS Speaker)" but in reality connected to "Comet Lake PCH-LP cAVS HDMI / DisplayPort 3 Output" (as shown by qpwgraph).

Setting it to "Comet Lake PCH-LP cAVS Speaker" made it work.

Click image for larger version

Name: Configure Read aloud_004.png
Views: 5
Size: 190.9 KB
ID: 217752

It’s unfortunate I now have to waste lots of space by having two sets of Piper and the TTS voices, but anyhow, thanks for making it easy for the rest of us, and chunking it so we can see nice highlighted reading progress. The direct voice download is also a nice feature for newbies.

Btw, can I simply throw my own ONNX voices (+ JSON) into the ~/.cache/calibre/piper-voices folder, or does it require more than that?

kovidgoyal · Yesterday, 10:40 AM

That should be enough and you dont need two copies of the voices you can use symlinks.

Yesterday, 04:44 AM	#1
Moonbase59 Addict Posts: 231 Karma: 1000244 Join Date: Oct 2021 Location: Germany Device: Tolino Vision 5, Tolino Tab 8", Pocketbook Era (16GB)	Calibre 8.9: TTS output, and where are TTS files stored? So I’m using Calibre 8.9 on EndeavourOS/Cinnamon (rolling Arch-based distro), and I’m experiencing some oddities with the viewer’s TTS output. I have a working speech-dispatcher output, using a selected set of Piper voices (English and German), and things like spd-say »This is a test.«, or Foliate work nicely. Now in Calibre’s ebook viewer, when selecting »Read Aloud«, it immediately started downloading a voice called libritts (United States) [High Quality]. I can also download other Piper voices. Where are these stored? I don’t really want to download & store all voices I use twice. When using the suggested The Piper Neural Engine, a reading progress is shown in the book (blue highlight) but I get no sound, whatever output device I select. When switching to my already-installed The Speech Dispatcher Engine, I get a list of my installed voices, can select one, and it reads the book text in the selected voice—but without any indication of position, and no page turning. Would be nice if this can be solved, preferably with the reading progress shown. Last edited by Moonbase59; Yesterday at 04:58 AM.

Yesterday, 10:19 AM	#3
Moonbase59 Addict Posts: 231 Karma: 1000244 Join Date: Oct 2021 Location: Germany Device: Tolino Vision 5, Tolino Tab 8", Pocketbook Era (16GB)	Yeah, I know, unfortunately that can’t work with SPD as it is. Thanks for the short insight to the pipeline. I’m using Pipewire, so I installed qpwgraph to check what happens, and surely enough, it had selected an output called "System default (currently Comet Lake PCH-LP cAVS Speaker)" but in reality connected to "Comet Lake PCH-LP cAVS HDMI / DisplayPort 3 Output" (as shown by qpwgraph). Setting it to "Comet Lake PCH-LP cAVS Speaker" made it work. It’s unfortunate I now have to waste lots of space by having two sets of Piper and the TTS voices, but anyhow, thanks for making it easy for the rest of us, and chunking it so we can see nice highlighted reading progress. The direct voice download is also a nice feature for newbies. Btw, can I simply throw my own ONNX voices (+ JSON) into the ~/.cache/calibre/piper-voices folder, or does it require more than that? Last edited by Moonbase59; Yesterday at 10:23 AM.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[GUI Plugin] TTS to MP3: Create MP3 audiobook using Windows TTS	jackie_w	Plugins	156	03-29-2025 11:11 AM
[Plugin] TTS - Windows TTS output plugin	Doitsu	Plugins	12	05-18-2018 09:43 PM
How to install "incompatible" apps to Boox M96+T68. IVONA TTS, ACAPELA TTS etc.	Booxtor	Onyx Boox	24	04-19-2017 05:08 AM
Problem with TTS on OnyxBoox M92 "invalid TTS data"	Marion Juin	Onyx Boox	1	03-27-2013 05:56 PM
GUI Launcher greyed out when TTS is on and does not come back when TTS is off	jmseight	Kindle Developer's Corner	4	07-28-2012 02:48 AM

Yesterday, 07:37 AM	#2
kovidgoyal creator of calibre Posts: 45,445 Karma: 27757438 Join Date: Oct 2006 Location: Mumbai, India Device: Various	If you want reading progress you have to use the piper engine. speech dispatcher doesnt have any. It's output goes via Qt Multimedia which in turn goes via ffmpeg which goes via pulse audio (or pipewire) which goes via the linux kernel to your hardware. something in that chain is broken in your system. Fix it and you will be fine.

Yesterday, 10:40 AM	#4
kovidgoyal creator of calibre Posts: 45,445 Karma: 27757438 Join Date: Oct 2006 Location: Mumbai, India Device: Various	That should be enough and you dont need two copies of the voices you can use symlinks.