Hi All: This message was forwarded to me by several people on this list. I'm in the process of figuring out how to join the list, so I'm using the web-based interface to compose my response right now.
I currently lead the Orca screen reader project and I also led the FreeTTS project. I think there are two separate problems here: 1) The quality of FreeTTS output when speaking the language it supports. 2) Which languages FreeTTS supports. The first is sometimes referred to as the "Swedish Chef" problem. I believe the joke came in around 1976 when Ray Kurzweil introduced his Reading Machine. At the time, people referred to the voice as sounding like the "Swedish Chef" from the Muppets. The second is what I think Lars is referring to. TTS engines need knowledge and support for the language of the text they are speaking. FreeTTS currently doesn't have support for Swedish as far as I know. Tossing Swedish text at it is like tossing Swedish text at a US English speaker and having them guess at the proper pronunciation. Hence, Lars' question about what is involved in making FreeTTS voices. Creating voices for FreeTTS is similar to creating voices for it's parent system, Festival. A lot of different kinds of data are needed, and it requires a fair amount of speech domain knowledge. In addition, given all this, you're never really guaranteed to end up with a good voice. As an alternative, I'm hoping we can port eSpeak (http://espeak.sourceforge.net/) and the gnome-speech driver for eSpeak to Solaris. I think it might be an easier task to accomplish this than creating a new voice for FreeTTS, and the benefits of this include the following: 1) eSpeak has support for a large number of locales, and the process for supporting new locales seems to be well documented. 2) eSpeak is smaller than FreeTTS, perhaps helping set us up for an accessible install. Lars - is working on a port of eSpeak to Solaris something you might be able to help with? Will This message posted from opensolaris.org
