Re: Read Etexts now supports Text To Speech
On Mon, Jun 9, 2008 at 9:09 AM, James Simmons [EMAIL PROTECTED] wrote: I know there are several people interested in having Text to Speech with Karaoke highlighting be a built in part of the Sugar environment. Also, when I originally requested a Git repository for the Read Etexts activity Ed asked if text to speech with highlighting would be supported. I was reluctant to commit to that at the time, thinking it would be too difficult. It turned out to be both easier and more difficult than I thought it would be, but I have released version 4 of the activity which now supports TTS with the words highlighted as they are spoken. Wonderful news. Now we have to talk to management about getting a project created to support more languages. Perhaps organized like Pootle, but with entirely different software. The idea is that we need * a linguistic analysis of the sound system of the language, or of any particular dialect * a script containing all of the sounds of the language for informants to read for recording * a process to create the files for the speech engine in the appropriate format * a dictionary and an orthography engine to convert from the written language to the required sound sequence. There will still be ambiguities that would require strong AI to resolve. He wound the bandage around the wound is a simple example of the problem. The code could be improved, no doubt. I am fairly new to Python programming. But I think trying out this Activity could give you some idea of what to expect if you attempt to incorporate TTS as part of the Sugar interface. 1). Speech-dispatcher needs to run in a separate thread from the GTK event loop, otherwise the callbacks needed to highlight words won't be received. 2). To get the callbacks as each word is spoken you need to format the text to be spoken as an XML document with tags *before* each word. My code assumes that words are separated by whitespace, which works for many languages but not all of them. I know Sanskrit doesn't work that way, for instance. Nor Chinese, nor Thai, nor a number of others. 3). Espeak does not allways do a callback for each word, and there is no obvious reason why any given word would be skipped. I understand that Festival works better, but I haven't tried it. At the suggestion of Hynek Hanke of the speech-dispatcher project I made the tag ids for each tag correspond to the word number in the document. In this way I can get the tag id in the callback and always highlight the correct word even if occasionally words are skipped over by espeak. 4). Pausing and resuming speech doesn't work. No idea why. 5). The instructions for setting up speech-dispatcher on the wiki are obsolete. You cannot use espeak-generic module with speech-dispatcher and get callbacks. You need to use the normal espeak module. When you try to use the normal espeak module with the current RPMs speech-dispatcher complains of a missing library. So if you want to try my Activity you'll need to use sugar-jhbuild with speech-dispatcher installed and configured to use espeak. Hemant Goyal is working on creating RPMs for speech-dispatcher and will be updating the instructions on the wiki. Is anybody interested in making the Debian/Ubuntu packages? This would be one of my favorite demos. The Activity page is: http://wiki.laptop.org/go/Read_Etexts James Simmons ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel -- Edward Cherlin End Poverty at a Profit by teaching children business http://www.EarthTreasury.org/ The best way to predict the future is to invent it.--Alan Kay ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: Read Etexts now supports Text To Speech
Edward, Ubuntu packages are already available for speech-dispatcher, espeak, and festival, so it isn't difficult to set up a demo. You can use apt-get. You would have to modify a configuration file to use the espeak module (not espeak-generic) and you would have to start speech-dispatcher by hand before doing the demo. There is a way to get it started automatically, but I never bothered to find out what it is. The instructions on the wiki are *close* to being right. Just substitute using the espeak module for espeak-generic and don't bother modifying the espeak config file. Hemant's RPMs will be specifically for the XO. Having special RPM's for the XO is desireable because we don't have a lot of disk space to work with and the usual package for speech-dispatcher brings in festival and other stuff that wouldn't be needed. This isn't a problem for ubuntu or debian running on a regular PC. I don't think current speech software is up to saying wound around the wound yet. I've tested with with Thuvia, Maid of Mars, Edison's Conquest of Mars, and Triplanetary and I've been impressed with how well the software deals with made-up words like you find in science fiction (Dusar, Ptarth), but I've also heard it pronounce micrometer as micro meter. For a kid learning to read I think it should be OK. But we have to have realistic expectations. I would love for the XO to sound like Hal 9000 (What do you think you're doing, Dave?) but I'll settle for Colossus The Forbin Project, which is pretty much what it sounds like. I've thought a bit more about highlighting for languages that don't split words. I convert my text one page at a time to SSML, using whitespace, but it should be possible to do just the oppposite: accept a document already in SSML and convert it to plain text for display. That way actual humans could figure out where the words split. Hemant was going to create a control panel for speech-dispatcher that allows you to set speech rate, voice used, etc. I was planning on using that so I didn't have to provide those options in my Activity. If that doesn't work out for some reason I was thinking of swiping one of the toolbars in the Speak activity and modifying that. Speech-dispatcher by default speaks very quickly, which I think is the preference for visually impaired people. I slowed it down quite a bit. Somehow or other this needs to be adjustable. Plus I need to get pause and resume working. James Simmons Edward Cherlin wrote: Hemant Goyal is working on creating RPMs for speech-dispatcher and will be updating the instructions on the wiki. Is anybody interested in making the Debian/Ubuntu packages? This would be one of my favorite demos. ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: Read Etexts now supports Text To Speech
Hi Edward, Ubuntu packages are already available for speech-dispatcher, espeak, and festival, so it isn't difficult to set up a demo. You can use apt-get. You would have to modify a configuration file to use the espeak module (not espeak-generic) and you would have to start speech-dispatcher by hand before doing the demo. There is a way to get it started automatically, but I never bothered to find out what it is. The instructions on the wiki are *close* to being right. Just substitute using the espeak module for espeak-generic and don't bother modifying the espeak config file. The debian package information is available here: http://packages.debian.org/stable/sound/speech-dispatcher OR http://www.freebsoft.org/pub/projects/speechd/snapshot-debian/ Hemant's RPMs will be specifically for the XO. Having special RPM's for the XO is desireable because we don't have a lot of disk space to work with and the usual package for speech-dispatcher brings in festival and other stuff that wouldn't be needed. This isn't a problem for ubuntu or debian running on a regular PC. Actually the RPM package at the moment is being made for the Fedora community in general. Once that is done, perhaps a special package with less features for OLPC can be made quite easily. Best, Hemant ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
