Cereproc TTS Voice - Problem

[Koz Myk]

I appear to be getting VA crashes when my new Cereproc Voice - Caitlin is being used.
It takes some time, maybe an hour or so, but I never had crashes with MS en-GB, Hazel.
In an evening's session I might get 2 or 3 VA crashes.

I've tested by going back to MS Hazel and no crashes ...

Anyone else had these kind of problems?

[Koz Myk]

Seems to have settled  more.
Only seems to happen when editing commands now, sometimes not every time.
Doesn't seem to like the Preview on Say Something with Text To Speech

[Gary]

Have you tried this with the latest beta?  It's 32 bit.

Gary

[Nate]

I recently purchased a cereproc voice and I am having similar problems. While editing, whether it's playing a voice or not, voice attack will freeze up/crash. When playing text to speech, sometimes the voice gets hung up on a slight delay on a word at the end of a sentence. Other times it's a huge delay of minutes and if multiple things are being said they stack on top of each other and once the delay is caught up it says everything all at once. 

Is there a way to know if this is a problem with VA or a problem with their voice? Ivona voices do not do this.

I am using the latest beta.

[Gary]

It's a problem with the voice with VoiceAttack.  I would suggest trying to use that voice with the beta, as it is 32 bit.  Also, might want to hit up the Cereproc support site and find out why every other tts voice seems to work ok with VA and not theirs.

Gary

[Nate]

I am on the latest beta, also I've contacted Cereproc and a couple people are investigating. 

[Gary]

Really weird...  So, you have Cereproc set as default in control panel?  I'm just going over the code and I don't instantiate any tts components until they are needed.  It must be affecting the speech engine in some way and is crapping out.  If you do NOT use TTS at all in a session, does it still crash?

Gary

[Nate]

I'll do some more testing and keep a log so I can give you details.

[Nate]

When using the cereproc voice and creating commands VA will crash within 10 minutes of use. VA does not crash when using Ivona voices. Here are some of the crashes I experienced:

• added Execute other command. Tried to re-edit, crashed/froze.
  
• While editing a command, tried to add Run windows app after a text-speech "say". Froze/crashed.
  
• Creating a new command, tried to add Run Windows app after a text-speech "say". Froze/crashed.
  
• New command, after creating a "Say" text to speech and previewing the voice I then tried to create a run windows app command. Froze/crashed.
  
• New command, tried to create "Say" text to speech in blank command. Froze crash.
  
• New command, tried to create "say" in otherwise blank command. Froze/crash.

It's very easy to reproduce. If I start VA and start editing commands even without ever using/hearing the TTS, it will still crash. 

When using Cereproc as the text to speech it will sometimes bunch up or delay the play back. No consistency to that though. 

On Saturday, April 25, 2015 at 4:18:46 PM UTC-4, Gary wrote:

[Gary]

This is only with CERTAIN Cereproc voices, right?

Gary

[Nate]

I'm using the Kirsty voice. I don't have any other voices to test.

[Matt Hingst]

Just posted something about this without seeing this thread. I have CereVoice - Jess  and it is crashing mostly on preview and use of the text to speech. Pretty sure it has crashed while just editing normal commands to though, as if just having her set as the default voice in win 7 is making voice attack crash.  

On this site:  https://www.cereproc.com/en/products/sdk

#include <cerevoice_eng_simp.h>
CPRCEN_engine * eng = CPRCEN_engine_load("license.lic", "speaker.voice");
CPRCEN_engine_speak_to_file(eng, "Hello there!", "out.wav");

there is an engine for rendering ! and ? , not sure if that has anything to do with it.

[Iain Lennon]

I'm using Cereproc Heather, and see the crashes in the release version. On the 1.5.8.67 Beta I find that occassionally the TTS is "stacked" for some reason: I have a command which speaks the time and the current record is a 12 minute delay! Once this has happened, further TTS often doesn't happen. Can be reset by dropping out and restarting and TTS then works again. Not sure this is cereproc or VA.

[Iain Lennon]

PS: Don't have the crashes in the beta though.

[Matt Hingst]

Odd I've still been having crashes in the beta. Just downloaded it to test to see if the crashes would stop. When using the beta I also got the "pile up" that didn't happen before using the beta.

On Sun, May 3, 2015 at 3:08 AM, Iain Lennon <iain.fam...@gmail.com> wrote:

PS: Don't have the crashes in the beta though.

--

---
You received this message because you are subscribed to a topic in the Google Groups "VoiceAttack" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/voiceattack/cpejDx2oKSQ/unsubscribe.
To unsubscribe from this group and all its topics, send an email to voiceattack...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Nate]

Has anyone tried the cereproc voices with a program other than Voice Attack? Do you have similar problems in those programs?

[Matt Hingst]

Ive used both natural reader and the default windows stuff and have yet to crash. Also switching from cerevoice to Ivona stopped the crashing (not ideal but works for now).

[Gary]

I'd like to be able to reproduce this.  I'm going to email Cereproc and see if they'll pony up a temp license for one of their voices.  

What is most perplexing is that where everybody describes having a problem, there is no tts synth initialized at all.  Basically, instantiate speech engine, speech engine does what it's been doing for the last five years... wait for crash.  I'm not entirely sure I'm going to be able to fix this o_O.  I'm going to be doing more work around the dictation stuff...  could be that doing it (however, dictation features do not exist in release version).  Will be looking into this.

Gary

[Gary]

Just realized something while working on a plugin...   I'm not sure when it happened, but the beta's have reverted to 'any CPU' and not x86.  This may or may not help (thinking it would not work at all if this was the case) but we can find out.  Expect a new beta today.   

Gary

On Sunday, May 3, 2015 at 6:26:51 PM UTC-5, Matt Hingst wrote:

[Gary]

Hey, Nate.  If you want, I put a new beta out today to see if it helps any.

Gary

On Sunday, May 3, 2015 at 2:35:33 PM UTC-5, Nate wrote:

[Nate]

It seems to have helped the text-to-speech playback as I have yet to hear the voice get caught or delayed as it's speaking. However if the cereproc voice is the default windows voice I will still get crashes when creating commands. So far I've gotten them when adding "execute another command", say something in text-to-speech, and run an application. But it is also not consistent. It's not like every time I add the command it crashes. It also doesn't appear to be based on whether or not the text-to-speech has already spoken or not. 

For whatever reason, it doesn't crash with Ivona voices as the windows default and I can still select the Cereproc voice in the command and it plays without error. As a work around I'm just selecting the voice per command. It would be great to be able to select a voice per profile. 

[Iain Lennon]

Slightly oddly, the beta has started crashing for me today. Have not made any significant changes to my set up (I use Windows only to play Elite Dangerous and live in a Linux world else).
Have also noticed that the cereproc TTS starts to stack when one command calls another, especially when they are stacked 3 deep, or 2 different functions are called from the same host function (I set up common ship functions as "subroutines" which more complex functions call). It's not absolutely reproducible though, and can still happen with simple commands.

am installing 1.5.7.69 as I type...

[Nate]

After using the latest update for a while, I will occasionally get a text-to-speech hiccup/delay, but it is nothing like the build up of delays I used to have.

On Tuesday, May 5, 2015 at 4:20:35 PM UTC-4, Gary wrote:

[Iain Lennon]

I'm still having big delays in the .69 release, to the point that the voices seize up, and commands don't complete waiting for them.

[Iain Lennon]

I've had a response from Cereproc today (I filed a bug report with them as well). They say they will be in touch...

[Gary Magenheimer]

I am in contact with their lead dev  I am going to be out of pocket for the next week, but after that I will be able to try to reproduce the problem in debug.  Nice guy over there.

Gary

On May 18, 2015, at 2:43 PM, Iain Lennon <iain.fam...@gmail.com> wrote:

I've had a response from Cereproc today (I filed a bug report with them as well). They say they will be in touch...

--

---
You received this message because you are subscribed to the Google Groups "VoiceAttack" group.
To unsubscribe from this group and stop receiving emails from it, send an email to voiceattack...@googlegroups.com.

[Gary]

Ok... I was finally able to get to this item for a bit.  The folks at Cereproc provided me a with a free license and I chose the, 'Heather (Scotland)' voice.  I will skip the boring details, but it boils down to the fact that Cereproc TTS voices can not be instantiated more than five times in a running application.  This was indicated by the Cereproc lead developer, and I confirmed it immediately ;)  VoiceAttack has synths that run out of a pool so you can (if you really wanted to) have multiple TTS voices running at the same time (and be able to manage the synths if you wanted to kill a command or whatever and have the tts stop talking).  Somehow, I figured that would be good for immersion (since now there's AI stuff out there with multiple voices), so, it's in there.  Anyway, since Cereproc doesn't like that, it stops responding after you use it five times (there would be up to five synths running).  Also, if you preview the voice five times (actually opening and closing the tts screen five times), it will freeze up the software (since a new synth instance is created each time the screen is loaded and the synth is needed).  To get around this, I added a new option to the, 'Options' page.  It is called, 'Single TTS Instance'.  It bypasses the pool and just uses a single, shared TTS synth, thereby staying under five instances.  It will be in the next beta release... no ETA, since I will be out of pocket for the next few days.

Gary

On Monday, May 18, 2015 at 2:57:47 PM UTC-5, Gary wrote:

I am in contact with their lead dev  I am going to be out of pocket for the next week, but after that I will be able to try to reproduce the problem in debug.  Nice guy over there.

Gary

On May 18, 2015, at 2:43 PM, Iain Lennon <iain.fam...@gmail.com> wrote:

I've had a response from Cereproc today (I filed a bug report with them as well). They say they will be in touch...

--

---
You received this message because you are subscribed to the Google Groups "VoiceAttack" group.

To unsubscribe from this group and stop receiving emails from it, send an email to voiceattack+unsubscribe@googlegroups.com.

[John Doe]

Great news that you got to the bottom of it. Looking forward to the update!

[Iain Lennon]

This is Much Better in .74. Single TTS has returned Cereproc Heather to me!

Only (slight) gripe is that now speech playing in a called, "nested" command stops any earlier speech in favour of the new commands words. This is a bit of a problem when reusing multiple commands in whole "state" changes as you lose the most early verbal feedbacks.

[Gary]

Yeah... unfortunately, due to the way Cereproc does their thing, it can only be single instance.  Everything else runs on multiple instance, but Cereproc...  no bueno.  The problem was that if I called multiple instances of TTS using Cereproc voices, it would cause VA (or any app) to crash.  This was confirmed with their (rather nice) lead developer who was willing to work with me on this. You might want to send them a message and encourage them to change their process and wait for it to happen, or, try another non-Cereproc TTS voice :(

Sorry about that!

Gary