From Ti@fa-kuan.muc.de Sun Feb 18 03:06:17 2001 Return-Path: X-Sender: Ti@fa-kuan.muc.de X-Apparently-To: lojban@yahoogroups.com Received: (EGP: mail-7_0_3); 18 Feb 2001 11:05:29 -0000 Received: (qmail 49111 invoked from network); 18 Feb 2001 11:05:29 -0000 Received: from unknown (10.1.10.26) by l7.egroups.com with QMQP; 18 Feb 2001 11:05:29 -0000 Received: from unknown (HELO mu.egroups.com) (10.1.1.40) by mta1 with SMTP; 18 Feb 2001 11:05:28 -0000 X-eGroups-Return: Ti@fa-kuan.muc.de Received: from [10.1.2.52] by mu.egroups.com with NNFMP; 18 Feb 2001 11:05:28 -0000 Date: Sun, 18 Feb 2001 11:05:28 -0000 To: lojban@yahoogroups.com Subject: Re: speech synthesizer Message-ID: <96oa9o+p5g0@eGroups.com> In-Reply-To: User-Agent: eGroups-EW/0.82 MIME-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Length: 1390 X-Mailer: eGroups Message Poster X-Originating-IP: 193.149.49.79 From: "A.W.T." X-Yahoo-Message-Num: 5520 --- In lojban@y..., Avital Oliver wrote: > On Fri, 16 Feb 2001, Arnt Richard Johansen wrote: > > [How to create a *real* Lojban speech synthesizer] > > > Anyone interested in the topic of high-quality speech synthesis might want > > to take a look at http://tcts.fpms.ac.be/synthesis/mbrola.html. > > Hmmm. I was at the site, and it looks *very* interesting. I think I'll try > starting to work on this... I'm thinking how we can build the corpus for > all the phonemes in Lojban. If anyone is intersted in helping/has > already done some work, inform me please. Was there too, and I think the diphone approach seems quite convincing in principle. Yet, to create a diphone database is not too easy a job and also pretty time consuming :( I also still do not know *exactly* how to accomplish this task. Listening to a couple of sound samples showed that others obviously didn't as well: the Hebrew one is far far too low, and the Romanian sample doesn't sound much Romanian at all, and - although listening to it several times - my wife and I didn't understand one single word! As for the (female voice) German sample (obviously(?) from Bavarian Television), the sound's distinct and clear, such that, initially, I didn't even have the idea it was *not* natural speech. So I think it all depends on the input's quality respective... co'o mi'e .aulun.