home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!ames!agate!doc.ic.ac.uk!sot-ecs!rid
- From: rid@ecs.soton.ac.uk (Bob Damper)
- Newsgroups: comp.speech
- Subject: Re: information about DECtalk
- Message-ID: <13565@ecs.soton.ac.uk>
- Date: 17 Nov 92 14:28:07 GMT
- References: <ulrike.721401334@gmd.de>
- Sender: news@ecs.soton.ac.uk
- Lines: 40
- Nntp-Posting-Host: donne
-
- In <ulrike.721401334@gmd.de> ulrike@gmd.de (Ulrike Bauer) writes:
-
- >I am a student who is writing her masters thesis about `generalization
- >in multilayer networks using the NETtalk files for training and
- >testing'.
- >
- >In this work I also compare some other systems generating phonemes
- >and stresses out of text, i.e. NETspeak, NETZSPRECH and DECtalk.
-
- Beware -- it is not a simple matter to compare text-to-phoneme
- systems. Even if they use the same principles (e.g.~both systems are
- based on back-propagation networks, like NETtalk and NETspeak), the
- testing reported by the authors is usually quite different -- to the
- extent that comparison is difficult. When the systems use totally
- different principles, it is not entirely clear how they should be
- compared. For a self-learning system like NETtalk, the tradition is to
- test on unseen data -- not present in the training set. But the
- concepts of `unseen data' and `training set' aren't even applicable to
- systems like DECtalk where rules are manually-derived by experts.
-
- I have written something on these problems -- the manuscript is
- nearing completion. Email me if you want to know more.
-
- >But till now I haven't got any real information about the DECtalk
- >system, how it works and which kind of research was done with it. I
- >haven't found any papers or articles concerning DECtalk.
-
- DECtalk is the commercial embodiment of the KLATTalk rule-based system
- from MIT. Major contributors were Dennis Klatt, Jonathan Allen and
- Sheri Hunnicutt. There's an enormous literature on text-to-phoneme
- conversion by rule: the above names figure prominently. Try the book
- ``From Text to Speech: the MITalk System'' by Allen, Hunnicutt and
- Klatt (1987).
-
- Bob Damper
- Vision, Speech and Signal Processing Research Group
- Department of Electronics and Computer Science
- University of Southampton
- Southampton S09 5NH
- United Kingdom Email: rid@uk.ac.soton.ecs
-