calibre 5.7.2: Voice selection

[DrChiper]

On windows 10 I have 4 default voices installed: George, Hazel, Suzan and Frank. I have selected per default Frank.
When I open the Read-out-Aloud setting in the viewer I miss the existing options George, Suzan and Frank. There seems to be an additional new voice named Zira however. (Suzan under another name?)
No matter which option I select in the viewer, I always seem to have selected Hazel (according to my limited audio recognition capabilities).
Do I miss some config settings to make the voice selection work?

[kovidgoyal]

voices come from the windows registry. David and Zira are the default voices in windows 10, with david being the system default. As for the rest whether they are recognized or not depends on where you go tht evoice from and whether the windows SAPI subsytem recognizes it.

[Uncle Robin]

Originally Posted by DrChiper

On windows 10 I have 4 default voices installed: George, Hazel, Suzan and Frank. I have selected per default Frank.
When I open the Read-out-Aloud setting in the viewer I miss the existing options George, Suzan and Frank. There seems to be an additional new voice named Zira however. (Suzan under another name?)
No matter which option I select in the viewer, I always seem to have selected Hazel (according to my limited audio recognition capabilities).
Do I miss some config settings to make the voice selection work?

This seems like the same issue I have in 5.7.1, as posted here. Additional MS voices installed from the available options listed in the Win 10 Control Panel, but not showing up as selectable in Calibre viewer.

[DrChiper]

Originally Posted by kovidgoyal

David and Zira are the default voices in windows 10, with david being the system default. As for the rest whether they are recognized or not depends on where you got the voice from and whether the windows SAPI subsytem recognizes it.

This is my default Windows configuration and I do not have added any additional voices since first installment. Also, I do not seem to have David which you mentioned though.

So perhaps calibre does not read the current registry setting correctly.
I'm using Windows 10 Pro 2004 build 19041.685

[kovidgoyal]

calibre doesnt read the registry, windows does. calibre just calls the windows SAPI SpEnumToken function to get the voices.

[jackie_w]

I wonder if there is anyone in the world who understands how Microsoft makes installed Voices available to various applications. The whole Windows registry is littered with lists of Voices, but which Voices appear in each list varies. Whoever thought this up must be some kind of sadist.

To the best of my knowledge (based on the work done on my 'TTS to MP3' plugin) the list of Voices calibre will see depends on whether you're using calibre 32-bit or calibre 64-bit. On my own Win10 PC:

calibre 64-bit can see all Voices listed under this registry key:

HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech\Voices\Tokens

calibre 32-bit (including calibre portable) can see all Voices listed under this registry key:

HKEY_LOCAL_MACHINE\SOFTWARE\WOW6432Node\Microsoft\SPEECH\Voices\Tokens

By default, Windows 10 Cortana Voices do not appear in either of these 2 lists. They can be added but it requires manual editing of the Windows registry, which is either brave or foolhardy depending on your POV. Here's some light(?) reading for the adventurous.

All of the above is based on observation/experiment rather than any real understanding.

[DrChiper]

Ok. In that case it seems that SpEnumToken has a preference for just the "Desktop" voice kinds. Which also explains why voice Zira is listed alongside voice Hazel.

[DrChiper]

Originally Posted by jackie_w

Whoever thought this up must be some kind of sadist.

Using Windows is a kind of self-torture I did gave up trying to understand Windows since version 3.11 ...

Originally Posted by jackie_w

Here's some light(?) reading for the adventurous.

Stumbled about that one too. It showed me how to find all narrators in settings.

In my case Win10 64x

HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech\Voices\Tokens

seems to have just the 2 desktop voices, and

HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech_OneCore\Voices\Tokens

seem to contain the 4 remaining voices

[JSWolf]

Originally Posted by jackie_w

I wonder if there is anyone in the world who understands how Microsoft makes installed Voices available to various applications. The whole Windows registry is littered with lists of Voices, but which Voices appear in each list varies. Whoever thought this up must be some kind of sadist.

To the best of my knowledge (based on the work done on my 'TTS to MP3' plugin) the list of Voices calibre will see depends on whether you're using calibre 32-bit or calibre 64-bit. On my own Win10 PC:

calibre 64-bit can see all Voices listed under this registry key:

HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech\Voices\Tokens

calibre 32-bit (including calibre portable) can see all Voices listed under this registry key:

HKEY_LOCAL_MACHINE\SOFTWARE\WOW6432Node\Microsoft\SPEECH\Voices\Tokens

By default, Windows 10 Cortana Voices do not appear in either of these 2 lists. They can be added but it requires manual editing of the Windows registry, which is either brave or foolhardy depending on your POV. Here's some light(?) reading for the adventurous.

All of the above is based on observation/experiment rather than any real understanding.

I have installed the registry entries to enable Eva and David.Eva does sound pretty good. David sounds like a badly built robot.

[jackie_w]

Originally Posted by JSWolf

I have installed the registry entries to enable Eva and David.Eva does sound pretty good. David sounds like a badly built robot.

Not too surprising. I think David may be a Windows 8 Voice whereas Eva is Windows 10 Cortana. There's another Win10 (not Cortana) US Male Voice, Mark, which should be accessible to those who have installed the US Voice pack. It may need registry editing shenanigans, though.

[JSWolf]

Originally Posted by jackie_w

Not too surprising. I think David may be a Windows 8 Voice whereas Eva is Windows 10 Cortana. There's another Win10 (not Cortana) US Male Voice, Mark, which should be accessible to those who have installed the US Voice pack. It may need registry editing shenanigans, though.

I just would like the TTS to read a little faster. But if I go up a speed, it reads to fast. So I want in between speeds.

If I had known David was rubbish, I would not have installed his reg entires.

[jackie_w]

Originally Posted by JSWolf

If I had known David was rubbish, I would not have installed his reg entires.

You can use regedit to delete the new David entry again if you want to. Of course, with the usual registry manual edit warnings.

[Uncle Robin]

Thanks for the regedit tips jackie_w! I have another problem with my voice selectrion now, though.

When I opened Calibre viewer yesterday, and opened the selection dialog for Read Aloud, it offered several "ScanSoft" voices. I selected one, but then found that it was not actually available, despite Calibre viewer ofering it as an option. Now, I can't access the selection dialog to change it to one of the voices made available by following the procedure above. I get this error message:

"OSError: C:\r\src\src\calibre\utils\windows\winsapi.cpp:163 :Failed to find voice with id:[-2147200966] IDispatch error #20026: HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech\Voice s\Tokens\ScanSoftTom_Dri20_22kHz"

It's a perfectly understandable error message, my problem is that it pops up when I click on the gear icon to try to change the selected voice to one that is available. I've tried a Repair of my Calibre install, but that didn't fix it. Any ideas on how I should proceed?

[jackie_w]

Originally Posted by Uncle Robin

Thanks for the regedit tips jackie_w! I have another problem with my voice selectrion now, though.

When I opened Calibre viewer yesterday, and opened the selection dialog for Read Aloud, it offered several "ScanSoft" voices. I selected one, but then found that it was not actually available, despite Calibre viewer ofering it as an option. Now, I can't access the selection dialog to change it to one of the voices made available by following the procedure above. I get this error message:

"OSError: C:\r\src\src\calibre\utils\windows\winsapi.cpp:163 :Failed to find voice with id:[-2147200966] IDispatch error #20026: HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech\Voice s\Tokens\ScanSoftTom_Dri20_22kHz"

It's a perfectly understandable error message, my problem is that it pops up when I click on the gear icon to try to change the selected voice to one that is available. I've tried a Repair of my Calibre install, but that didn't fix it. Any ideas on how I should proceed?

Info dump incoming ...

I also own several old purchased ScanSoft Voices, including Tom. I think they are Win 32-bit-only Voices. I can only see mine when I use calibre portable, which is 32-bit and, as I said earlier in the thread, I think the 32-bit Voice list is coming from registry key

HKEY_LOCAL_MACHINE\SOFTWARE\WOW6432Node\Microsoft\SPEECH\Voices\Tokens

The registry key in your error message shows the key I associate with calibre 64-bit. I don't think any of the ScanSoft Voices belong under that key. Caveat, this may be specific to my Scansoft Voices which were originally bought (circa 2007) for use with Nextup TextAloud which is a 32-bit program.

This is the current contents of my registry entry for ScanSoft Tom. This is what the Tom installer exe put in my registry.

Spoiler:

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SOFTWARE\WOW6432Node\Microsoft\SPEECH\Voices\Tokens\ScanSoftTom_Full_22kHz]
"LangCode"="ENU"
"Language"="American English"
"Frequency"="22"
"Encoding"="155mrf22"
"Reduction"="FULL"
"VoiceName"="Tom"
"VoiceData"="C:\\Program Files (x86)\\NextUp-ScanSoft\\speech\\components\\data\\Tom.dat"
"CLSID"="{CAC6785B-655E-4ae1-A656-BDEFD18DC46C}"
@="ScanSoft Tom_Full_22kHz"
"409"="ScanSoft Tom_Full_22kHz"

[HKEY_LOCAL_MACHINE\SOFTWARE\WOW6432Node\Microsoft\SPEECH\Voices\Tokens\ScanSoftTom_Full_22kHz\attributes]
"Vendor"="ScanSoft, Inc"
"Name"="ScanSoft Tom_Full_22kHz"
"Language"="409"
"Gender"="Male"

[Uncle Robin]

Thanks Jackie. I think I may have been a bit imprecise about the nature of my problem. Those scansoft voices appeared yesterday in my Calibre viewer's "Configure Read Aloud" list of voices. They were not showing up in my Calibre list when I posted these screenshots two days ago, then they appeared, and I stupidly selected one. The error message I posted above is 100% accurate, because a regedit search shows no entries in any key in the entire registry for any ScanSoft voices. My problem is that I can't access the Calibre viewer's "Configure Read Aloud" dialog to repalce the missing voice with one I do have, because attempting to do so by clicking the gears icon brings up that error message.