|
From: | David Hill |
Subject: | Re: [gnuspeech-contact] GNUstep version |
Date: | Tue, 13 Jan 2009 13:13:56 -0800 |
Hi everyone, Just a few comments on this exchange. My own (hopefully informed) opinion, of course, for which only I am responsible. I will also take this opportunity to ask if anyone on the list is willing help with the work we are doing. On Jan 13, 2009, at 12:10 PM, Marcelo Yassunori Matuda wrote:
"Gnuspeech" is really a *suite* of programs, as described on the savannah web page for the project: Monet is probably the most important and inclusive part because: (a) it allows the articulatory speech synthesis to be checked out with arbitrary short input text segments; (b) it forms the core of the tools needed to build databases for an arbitrary language; and (c) Monet involves nearly all the components needed for other items such as the TextToSpeech (TTS) server, or daemon, that allows unrestricted text-to-speech synthesis as a service to any application. For the OS X stream, this TTS server daemon has just been ported/rewritten by Dalmazio Brisinda (who gained significant experience with the original NeXT suite whilst implementing an audio-tactile system to allow visually disabled users to access documents using many cues translated from the visual domain into auditory and tactile form). The daemon it is basically Monet with all the interactive and other unnecessary stuff stripped out, plus other changes appropriate to its revised purpose. The complete suite of Apps comprising the nascent Gnuspeech is described in the web page cited above, which also indicates some history and includes the main originators. Monet was ported/rewritten for Mac OS X from the original NeXT implementation by Steve Nygard who also did some work on the original NeXT version. For Marcelo to have got a working (if buggy/untested) version of Monet working on GNUstep, is a very significant advance (thanks Marcelo) towards our goal of having the whole Gnuspeech suite running under GNUstep on GNU/Linux. The OS X work has been of significant help in this endeavour because of the very close similarities to NeXTSTEP, OpenStep, GNUstep, coupled with being more complete much sooner than GNUstep. I am currently rewriting/porting "Synthesizer", to OS X as a step towards a GNUstep version. Synthesizer is the second important component in creating the databases needed for arbitrary language articulatory text-to-speech synthesis -- in the absence of a comprehensive set of articulatory X-rays of the vocal tract making the sounds of the language, since it allows the sounds of a language to be related to the articulatory postures needed for articulatory synthesis, as well as giving the user insight into the vocal tract model. The third component (and not as easy as it may sound) is the creation of the necessary pronunciation dictionary (plus letter-to-sound rules to catch words not in the dictionary). As part of the third component, a parser is also necessary to manage derivatives of dictionary words, handling of special cases (e.g. dates, real numbers, abbreviations and such). These third component items already exist for English and are incorporated (or some are soon to be incorporated, since there are several layers) in all of the NeXT, OS X and GNUstep versions of Monet. The dictionary has more than 70,000 primary words in it, and the parser manages derivatives. "Synthesizer" is close to being available in a 0.5 version release.
There is the manual for the OS X version of Monet. It would be fairly straightforward to replace all the screen shots with the corresponding NeXT version screen shots. Using real GNUstep screen shots might take a little longer.
Maybe so far. Lots of alligators in my swamp but I hop to do some testing myself shortly. "Synthesizer" is my top priority for now
All good wishes. david -------- David Hill Gnuspeech project admin -------- Simplicity, patience, compassion. These three are your greatest treasures (Tao Te Ching #67) --------- |
[Prev in Thread] | Current Thread | [Next in Thread] |