We are pleased to inform you that we have made some progress with getting Festival TTS going for ROILA. As always feedback and comments are welcome. Please note that this is still work in progress.
Here are the initial results (for sake of consistency, we provide audio files of sample sentences that are also discussed elsewhere)
fosit koloke
fosit nole
buse fosit
bama buse fosit
bobuja
fosit jimeja
fosit kipupi
fosit webufo
fosit besati
The output from text to speech was also passed to the Sphinx-4 speech recognizer and the recognition results were good.
Several steps are required to accomplish TTS for ROILA and ultimately to get your audio files. We will post details very soon so that everyone can have their machines talk in ROILA!
Acknowledgement: We would like to thank Steve Pomeroy for providing a sample script of how to enable Festival to speak in artificial languages. We would also like to thank Alex Juarez for helping in running Festival on Linux.
Tags: festival, text to speech
-
Although it’s not cheap and technically directly usable from a “ROILA text generator” in this form, you could also try to use the Acapella demo site to try out the pronunciation on different human accents, as this techno produce something much more human-like than your first results with Festival.
http://www.acapela-group.com/text-to-speech-interactive-demo.htmlI tried a bit the first examples and I found a few voices having a more or less close pronunciation to the one you got with Festival. Especially the Spanish (and American Spanish) voice, the Italian one, Czech, Finnish, German, Turkish and Norwegian and Swedish a bit.
“bobuja” seems to be problematic for most of them… (I stopped after this example).This raised me a question. If you intend to make a language that is more understandable to robot, how are you going to go through the regional accent distortions? Because some of the other voices were really pronouncing some syllabus in a different ways!
I made this “investigation” rather quickly. I don’t have time to fully explore every examples with every voices given in demo by Acapella. Plus sometimes their application is not really responding…
I guess you could also try with some other TTS provider to see how the regional pronunciation could impact the design of the language. -
The “t” in end of word “fosit” is almost silent.
It’s hardly recognizable and not noticeable in “fosit jimeja”.
Is that right?P.S. Does pronunciation indicate the word boundaries?
-
Are you trying to make speech synthesizers pronounce ROILA and make them think it’s Spanish/Finnish/German/Turkish/Norwegian/Swedish/… ?
Festival should provide something to generate speech from PHONEMES, not letters. English/some other language has all these phonemes, just use the voice for such language, and turn ROILA letters to phonemes by yourself and then pass phonemes to Festival. -
This is almost what we done. We use an English voice to pronounce the ROILA phonemes.
-
Guys, any update? I kind of need ROILA for a research project. It would be great if we had even a partially working synthesizer.
-
Dear Kadir,
The Sphinx based synthesizer is working perfectly. Please contact Omar if you have any question on the setup.
6 comments
Comments feed for this article
Trackback link: https://roila.org/2010/07/27/first-attempts-with-festival-for-text-to-speech-for-roila/trackback/