calibre friday preview releases - Page 6

kovidgoyal · Yesterday, 11:46 AM

This should take care of it:
https://github.com/kovidgoyal/calibr...d9760f8d60a963

Let me know how it goes for you.

Comfy.n · Yesterday, 11:59 AM

yes, that fixed it

thiago.eec · Yesterday, 01:00 PM

Hi, Kovid

Thanks for the new TTS engine. I was having the same problems as @Comfy.n. Now they are fixed.

I noticed another glitch: every other sentence, the last word is not being highlighted. I couldn't find a pattern. Also the Piper voices are not available in my Content Server (using Chrome). I can see Chrome voices and Windows, but not Piper.

Now, two questions:

1) I don't see how I can manage Piper voices data. I mean, how can I delete a voice, if I don't like it?
2) This is not something new, but I just remembered it. Would be possible to have an option to automatically choose the TTS language based on the <dc:language> element for EPUBs (and the equivalent for other formats)?

kovidgoyal · Yesterday, 01:18 PM

The content server cant use piper, it can only use whatever voices are available on the device the browser is running on. Piper is *very* CPU intensive I don't think its appropriate to expose it via the server as doing so could lead to the server being DOSed very easily. Just two people using it at the same time would saturate the CPU of most servers.

If a sentence is too long it is split up that is probably the cause of your last word not being selected, this is done so that audio starts playing asap.

1) There is no UI for it, but you can just go in to the calibre cache directory and delete the piper folder or the voice .onnx files inside it, they have proper filenames.

2) The voice is chosen based on calibre UI language not book language. I'm not really a fan of having the voice changed automatically based on book language, but maybe an option for it. SOmeday if I am bored, or a patch is welcome.

thiago.eec · Yesterday, 03:50 PM

Quote:

Originally Posted by kovidgoyal

Piper is *very* CPU intensive I don't think its appropriate to expose it via the server as doing so could lead to the server being DOSed very easily. Just two people using it at the same time would saturate the CPU of most servers.

I was monitoring the CPU activity and that's what I saw: there is a peak in CPU usage during about 10-20 seconds (in my case, piper.exe was using 50% of my CPU), then it drops to 0% and stays there. Anytime you click on a new paragraph, pause, or make any configuration change, the peak happens again, but after that it always seat at 0%. Is that expected or maybe the peak repeat itself after sometime (something related to the cache size)?

Quote:

Originally Posted by kovidgoyal

If a sentence is too long it is split up that is probably the cause of your last word not being selected, this is done so that audio starts playing asap.

The weird part is that it is always the last word, never more. It happens even with quite small sentences (10 words or so).

Quote:

Originally Posted by kovidgoyal

1) There is no UI for it, but you can just go in to the calibre cache directory and delete the piper folder or the voice .onnx files inside it, they have proper filenames.

I think it would be good if there was a button to clear a downloaded voice.

Quote:

Originally Posted by kovidgoyal

2) The voice is chosen based on calibre UI language not book language. I'm not really a fan of having the voice changed automatically based on book language, but maybe an option for it. SOmeday if I am bored, or a patch is welcome.

Yes, an option on the dropdown menu would be the best approach, so it is not forced on anyone. I hope you get bored soon enough.

kovidgoyal · Yesterday, 10:25 PM

CPU usage will spike when speech starts, that is when the text is being synthesized into audio, the synthesis happens faster than the actual speech (otherwise you would get pauses after every sentence), so once all the text for the current file is synthesized it will idle again.

I cannot reproduce the last word not being highlighted from sentences. If its happening on a particular book open a bug report and attach the book.

Added a button to download/remove voices: https://github.com/kovidgoyal/calibr...1aacbba0a7f8e0

Yesterday, 10:25 PM	#81
kovidgoyal creator of calibre Posts: 44,286 Karma: 23661992 Join Date: Oct 2006 Location: Mumbai, India Device: Various	CPU usage will spike when speech starts, that is when the text is being synthesized into audio, the synthesis happens faster than the actual speech (otherwise you would get pauses after every sentence), so once all the text for the current file is synthesized it will idle again. I cannot reproduce the last word not being highlighted from sentences. If its happening on a particular book open a bug report and attach the book. Added a button to download/remove voices: https://github.com/kovidgoyal/calibr...1aacbba0a7f8e0 Last edited by kovidgoyal; Yesterday at 11:45 PM.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Dockerise multiple Calibre releases for plugin development	wolf123	Related Tools	1	02-06-2021 03:12 AM
[Q] How can I check for new releases in a series in calibre or goodreads?	lambstone	Kobo Reader	1	09-09-2016 12:04 AM
CC B4.0.X beta (all releases)	chaley	Calibre Companion	41	07-21-2015 06:53 AM
CC beta releases 3.4.4j, 3.4.4k, 3.4.4l	chaley	Calibre Companion	33	10-21-2014 06:19 AM
dotReader releases first public beta	Alexander Turcic	Reading and Management	4	11-23-2006 01:40 AM

Yesterday, 11:46 AM	#76
kovidgoyal creator of calibre Posts: 44,286 Karma: 23661992 Join Date: Oct 2006 Location: Mumbai, India Device: Various	This should take care of it: https://github.com/kovidgoyal/calibr...d9760f8d60a963 Let me know how it goes for you.

Yesterday, 11:59 AM	#77
Comfy.n want to learn what I want Posts: 1,113 Karma: 6425162 Join Date: Sep 2020 Device: Calibre E-book viewer	yes, that fixed it

Yesterday, 01:00 PM	#78
thiago.eec Guru Posts: 992 Karma: 1183425 Join Date: Dec 2016 Location: Goiânia - Brazil Device: iPad, Kindle Paperwhite	Hi, Kovid Thanks for the new TTS engine. I was having the same problems as @Comfy.n. Now they are fixed. I noticed another glitch: every other sentence, the last word is not being highlighted. I couldn't find a pattern. Also the Piper voices are not available in my Content Server (using Chrome). I can see Chrome voices and Windows, but not Piper. Now, two questions: 1) I don't see how I can manage Piper voices data. I mean, how can I delete a voice, if I don't like it? 2) This is not something new, but I just remembered it. Would be possible to have an option to automatically choose the TTS language based on the <dc:language> element for EPUBs (and the equivalent for other formats)?

Yesterday, 01:18 PM	#79
kovidgoyal creator of calibre Posts: 44,286 Karma: 23661992 Join Date: Oct 2006 Location: Mumbai, India Device: Various	The content server cant use piper, it can only use whatever voices are available on the device the browser is running on. Piper is very CPU intensive I don't think its appropriate to expose it via the server as doing so could lead to the server being DOSed very easily. Just two people using it at the same time would saturate the CPU of most servers. If a sentence is too long it is split up that is probably the cause of your last word not being selected, this is done so that audio starts playing asap. 1) There is no UI for it, but you can just go in to the calibre cache directory and delete the piper folder or the voice .onnx files inside it, they have proper filenames. 2) The voice is chosen based on calibre UI language not book language. I'm not really a fan of having the voice changed automatically based on book language, but maybe an option for it. SOmeday if I am bored, or a patch is welcome.