A tool to generate voice files using various TTS providers

[Caffeine]

Hy,

ERROR: this is not a 32bit version only 64bit ??

I am on WIN7 32bit

It's built to target both 32 and 64 bit.

Did you install Microsoft .Net 4.5.2 as per the first post?

[jhsa]

I have 4.5.51, I guess that is ok as it is more recent?

João

[Caffeine]

Yes, that's fine.

You get the 'Not a valid Win32 Application' if you don't have the framework installed.

[jhsa]

I have just created a pack for me, and I'm at the moment normalizing the volume in audacity as the files come out of the utility at a lower volume..

João

[Caffeine]

The Google interface doesn't have a volume setting. The Microsoft TTS engines do, and it's set to maximum.

The playback mechanism within the tool also has a playback volume, and they it also set to maximum.

[jhsa]

Not a big problem.. it is very easy to normalize the volume in audacity. Just have to be careful to save them in the same sample rate and bit rate..
I did it to my pack last night..
It would be great , if we could set the speed of the speech for a given line, or for all, as someone else requested.. one less thing to do in audacity
By the way, I have created lots of new files last night and found the interface very easy to work with.. Really nice..

Thank you.

João

[Caffeine]

Added volume and speech rate control for local TTS engines.

TTSTranslate 1.5.1.0.zip

(490.59 KiB) Downloaded 217 times

[Caffeine]

Added functionality to insert rows above or below the currently selected rows.

It gets a bit confused when you select by dragging up rather than down, so I'll investigate that further.

TTSTranslate 1.5.2.0.zip

(491.34 KiB) Downloaded 200 times

[jhsa]

Thank you. Will test in a while..

João

[jhsa]

Move line/selection up and move line/selection down would also be a good addition. I just tried to organize my pack and had to insert a line, then cut and paste all the cells's contents to the new created line..

Couldn't find any funny behavior by dragging up or down to select lines

Thanks

João

[Caffeine]

Interesting feature. I'll try to add that tonight. Thanks for the feedback!

[jhsa]

What about this?? Can you use your utility to generate files from here? the voices seem to be quite decent as well..

http://www.fromtexttospeech.com/

Thanks
João

[Caffeine]

It looks interesting. I'll have a look at their interface and see if I can integrate it. It looks quite slow though.

[jhsa]

Try the US-Alice and the UK-Emma voice..
Emma's voice reminds me of Dana Scully's voice in the X-files

If you manage to integrate it, your utility wouldn't depend on google only.. Also it seems that this site allows for selecting voice and speed

João

[Caffeine]

I've implemented a quick attempt at using http://www.fromtexttospeech.com. It works but it's pretty fragile, and very slow. Their site redirects to a page where you can download the mp3, which I then scrape to find the filename. Often their site times out.

In comparison, Google directly returns the mp3.

I cleaned up the code to select between the different TTS providers, but it's still pretty ugly. I'm going to have to refactor a lot of it soon.

So in summary:

• Google interface : Medium speed, great sounding voices, might stop working without warning
• Microsoft TTS engine : Super fast, OK sounding voices, will not stop working. No downloads required
• FromTextToSpeech.com : Incredibly slow, fragile, good sounding voices, might stop working without warning

TTSTranslate 1.5.3.0.zip

(493.21 KiB) Downloaded 187 times

[Caffeine]

I haven't implemented an interface for speed setting yet.

[jhsa]

thanks, will try now..
João

[jhsa]

Another idea for you
When we make a pack, at least I tend to go through all the files and listen, to check if some needs changing.. What about if we could mark them for changing and come back to edit later? I guess it doesn't make sense saving the markings with the phrase file as when we load a new one the program wants to build the pack again? It would be nice also that it remembered that it build the pack already, and could play the files from the folder.. I guess this might be difficult to implement?
Thanks
João

[jhsa]

One thing I notice, is that the voices have more quality when played on their internet site.. Specially the "S" at the end of the word.

João

[Caffeine]

That could be because of the mp3 to wav conversion. You're playing the mp3 on their site, vs the converted wav in the tool. Does the downloaded mp3 sound the same?

Also, if appears that the fromtexttospeech page uses Ivona TTS so I may be able to get the rest of the voices working.

looks like an Ivona developer account allows 50000 units of TTS a month for free. A unit is up to 200 characters, so each of the phrases would be a single unit.

[jhsa]

Nice.. Will check the mp3 to see if it sound better.

Forgot to say that I built a pack with the voice Emma. You were right, it was slow. Had one time out..
After the timeout, even after pressing stop and then "Go" again, it didn't start the download.. I had to restart the program and delete the files that were already done..
This is where the program recognizing the files that are already in the folder would come very handy

João

[jhsa]

Just compared the mp3 and the wav files.. There is a difference, and it is quite noticeable. I think the problem is on the conversion after the download.

João

[Caffeine]

OK, I'll have a look at it.

[jhsa]

Please check the attached file.
There are 2 files.
The file "Easystar" was converted by the utility.
The file "Easystar_1" was converted in Audacity.

The mp3 files sample rate is 22050Hz, about 22Khz.
Also, is it possible to eliminate the metadata from the files automatically? Metadata is not good for our application and it can cause problems.

Thanks
João

EDIT: It looks like the utility is cleaning the metadata already

[Caffeine]

I have generated a full set using the fromtexttospeech interface, and I'll investigate now why the conversion to mp3 is degrading the quality.

I couldn't detect a difference when I compared the Google voice files, as they go through the same conversion process.

[jhsa]

You're right, the google mp3 and wav files seem to sound the same, although they have some other noises
Some kind of digital noise.. The others don't seem to suffer from it..

João

[Caffeine]

Converting the fromtexttospeech files from 22050Hz MP3 to 22050Hz wav files using the same method in the tool produces files that sound extremely similar. It appears to be the downsampling to 16kHz that's causing the issue.

Here's an example;

S.zip

(239.77 KiB) Downloaded 168 times

[Caffeine]

I have the full Ivona library working. I'll now integrate it in to the tool and remove the fromtexttospeech option (or move it to the end in case I use up all my Ivona credits in a month.

That means 51 voices

[jhsa]

It's not only you, it's 51 voices for everyone that might use the tool
Yes, please leave the other option in. It takes nearly an hour but it works.
Can anything be done about the downsample problem?
For me is not a biggie as it is quite fast to do it in audacity, but many might not know how to do it.
Thanks

João

[Caffeine]

Here we have Ivona TTS integration. I've set it as the default because it sounds much nicer than Google, and is almost as fast.

Since we're not exclusively using Google Translate, I've changed the name of the tool to TTSAutomate.

As the developer licence for Ivona permits usage solely for Developer’s internal Product development and not for any commercial or other purpose, you are all officially my software test team.

TTSAutomate 1.6.0.0.zip

(498.58 KiB) Downloaded 147 times