Dictionary

[jack Schmidling]

First problem that I have run into is the speed control.

Using Mike16, 0 is much slower than -1 and +1 is too fast. I can't seem to get the speed I want.

Also, the test speed does not seem to agree with the actual run speed.

Any ideas?

Jack

[Jim Bretti]

Hi Jack,

We use the Sapi5 speech interface in TextAloud, and Sapi5 specifies how the scaling on the speed control slider should look. So there is no way for us to fine tune the Speed ... hopefully future versions of the Sapi speech interface will provide this. One possibility may be to switch TextAloud to use the older Sapi4 speech interface. Go to the TextAloud voice options panel (Tools -> Voice Options) and change the Sapi Version Dropdown to "Use Sapi4 Only". The Sapi4 speech interface has a completely different spec for voice speeds, and you may be able to get the speed you're looking for. You'll lose some functionality specific to Sapi5 ... mainly the tags under the Insert Tags menu. You'll still be able insert pauses, but the speed, pitch, volume, spell, and voice change tags will not work.

I can't explain why your test speeds seem to be off ... I'll do some checking here and post something if I find anything.

[jack Schmidling]

Hi Jack,

You'll lose some functionality specific to Sapi5 ... mainly the tags under the Insert Tags menu. You'll still be able insert pauses, but the speed, pitch, volume, spell, and voice change tags will not work.

You lost me there. What is the point if the speed change will not work? Perhaps because I do not know what the tags are.

For the record, my only interest is in generating audio versions of books that are not available in audio. These are one time reads with no changes other than global fixes of pronunciation using the dictionary.

I am pretty happy with how this is working out so far but don't want to leave any stones unturned.

Jack

[Jim Bretti]

Hi Jack,

Sorry about the confusion, I didn't explain that very well. The speed tags under the Insert Tags menu can be used to dynamically speed up and slow down as text is being spoken. These tags have no effect on the default speed assigned to a voice. So for example, with a Sapi5 voice, you can set the default speed to 1. Then, anywhere inside the text, you can use Insert Tags -> Absolute Speed to change the speed to 2. When this tag is reached while text is being spoken, the speed changes to 2, without affecting the default speed for the voice.

These tags are ignored when a Sapi4 voice is assigned to the article.

[jack Schmidling]

Ok so that does not apply to my application as I just want a one time read and won't be tagging anything.

After reading the thread on favorite voices, I downloaded Lawrence Cepstral and find that many of the dictionary changes are either unnecessary or don't work. Am I supposed to create a separate dictionary for each voice?

Also, as many of these threads are very old, I hate to buy another voice and find that there is something better. I am looking for a cultured British voice and this is pretty close.

js

[Jim Bretti]

If you visit our voices page at http://www.nextup.com/TextAloud/SpeechE ... oices.html you'll find quite a few British voices, from AT&T, Nuance and Acapela.

On your question about whether or not pronunciation changes are voice specific, you can configure this either way when you create/edit dictionaries. If you edit the dictionary containing your pronunciation changes, you'll see a section at the bottom of the edit dialog labeled "Dictionary Associations". You have the option of making pronunciation changes in the dictionary specific to a language / vendor / voice.

Let me know if that doesn't answer your question.

[jack Schmidling]

Thanks.

Now for another question that just came up leaving me sort of in awe of at&t.

When either att voices speak the following:

It became clearly split into two bright but minute suns, but a minute later

the pronunciation of minute is correct in each context.

When Cepstral Lawrence speaks it, he gets it exactly backwards.

I was about to purchase Lawrence but this is a real show stopper.

I don't understand how any computer could get this right but maybe it's just a fluke.

BTW, switching to sepi4 allows a nice fine tuned speed control as discussed before but does not allow playing of the Cepstral voice. Any idea why this would be?

js

[Jim Bretti]

Hi Jack,

Sorry, but I forgot that your problems started with Sapi5 speed scaling, and we were trying to use Sapi4 instead. The only voice we have available that works with both Sapi4 and Sapi5 is AT&T. To get Cepstral to work in TextAloud, you will need to go back to the Voice Options panel and set the Sapi Version dropdown to "Use Both". This should get Cepstral working, but you'll need to see if the speed scaling is still an issue with the voice.

If you'd like, send an email to me at jim@nextup.com, and I'll send you an mp3 of Acapela Peter speaking your sample sentence at speed 0.

[PHenry1026]

Greetings:

It became clearly split into two bright but minute suns, but a minute later

TextAloud has a big problem with pronouncing homographs like (live, read, minute, etc.). If the above problem you encountered occurs in an AT&T NV voice it can be easily fixed (see the following post: http://www.nextup.com/phpBB2/viewtopic.php?f=13&t=6184). However, there is no adequate fix for this problem if it occurs in Lawrence Cepstral or any of the non-AT&T voices. I also like and have Lawrence Cepstral but it (and other non-AT&T voices) is basically useless without the Specialized Phoneme Dictionary I have proposed (see the following post: http://www.nextup.com/phpBB2/viewtopic.php?f=13&t=6184).

A Specialized Phoneme Dictionary would provide an easy fix to the homograph problem for non-AT&T voices.

Percy
RC Dictionaries

[jack Schmidling]

Greetings:

It became clearly split into two bright but minute suns, but a minute later

TextAloud has a big problem with pronouncing homographs like (live, read, minute, etc.). If the above problem you encountered occurs in an AT&T NV voice it can be easily fixed (see the following post: http://www.nextup.com/phpBB2/viewtopic.php?f=13&t=6184).

I guess I don't understand this but one can already tell it to pronounce live as liv or possible lyve but the dictionary makes global changes, i.e. will always pronounce it that way till changed.

I do not understand how NV got the pronunciation of minute correct in my above example.

Thanks for the new word. (homograph)

js






However, there is no adequate fix for this problem if it occurs in Lawrence Cepstral or any of the non-AT&T voices. I also like and have Lawrence Cepstral but it (and other non-AT&T voices) is basically useless without the Specialized Phoneme Dictionary I have proposed (see the following post: http://www.nextup.com/phpBB2/viewtopic.php?f=13&t=6184).

A Specialized Phoneme Dictionary would provide an easy fix to the homograph problem for non-AT&T voices.

Percy Henry

[PHenry1026]

Greetings again,

It was just mere luck that AT&T NV pronounced minute correctly in this case; to force it to pronounce minute correctly you would need a regular expression like the following:

(?#mynute)(?m)(?<=^|\s|['"‘“(]|\p{Pi}|\p{Ps}|\p{Pd})(?<=split into two bright but )minute(?= suns)(?=[’”\p{Po}\p{Pe}\p{Pf}]{0,2}(?:\s|$))

mynute

The above example assumes that the respell (mynute, the proxy) has been correctly phonetically defined in AT&T NV dictionary.

Using liv or possible lyve to pronounce live is a very poor approximation that can vary over different examples. You can get 100% accurate pronunciations of both possibility for live by defining phonetic proxies for the possibilities in AT&T NV Dictionary.

If TextAloud implements the Specialized Phoneme Dictionary I discussed above non-AT&T voices would also have this possibility.

Percy
RC Dictionaries

[jack Schmidling]

Greetings again,

It was just mere luck that AT&T NV pronounced minute correctly in this case; to force it to pronounce minute correctly you would need a regular expression like the following:

(?#mynute)(?m)(?<=^|\s|['"‘“(]|\p{Pi}|\p{Ps}|\p{Pd})(?<=split into two bright but )minute(?= suns)(?=[’”\p{Po}\p{Pe}\p{Pf}]{0,2}(?:\s|$))

You are way ahead of me as I haven't clue what any of that is or where it goes.

To avoid learning what I don't need, let me restate my objective.

I want to convert books to audio that are not already available in that form. This is for my own use only and for a one time read. Any changes I make would be on a global basis such as changing M. to mis your or valet to valay etc.

I can live with only using NV voices.

It appears that your code above would be inserted into a specific text to fix it but I am not sure about that.

If it will always make it right some how, I am interested. If it only relates to a specific text in a specific book, it's not worth my trouble.

Thanks,

js



mynute

The above example assumes that the respell (mynute) has been correctly phonetically defined in AT&T NV dictionary.

Using liv or possible lyve to pronounce live is a very poor approximation that can vary over different examples. You can get 100% accurate pronunciations of both possibility for live by defining phonetic proxies for the possibilities in AT&T NV Dictionary.

If TextAloud implements the Specialized Phoneme Dictionary I discussed above non-AT&T voices would also have this possibility.

Percy Henry

[jack Schmidling]

One more issue related to pronunciation is alluding me.

Mike and Crystal frequently accent the wrong syllable but I can not find a clue as to how to change this in the dictionary.

Surely, this simple problem has a simple fix but I see nothing in the help list on the subject.

Jack

[PHenry1026]

Greetings Jack,

I believe the following post addresses your question: http://www.nextup.com/phpBB2/viewtopic.php?f=13&t=6296

Percy
RC Dictionaries