Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!watmath!clyde!burl!ulysses!allegra!princeton!orsvax1!pyrnj!caip!daemon From: local-info-amiga-request@ics.UCI.EDU Newsgroups: net.micro.amiga Subject: Re: AmigaTerm with voice Message-ID: <2003@caip.RUTGERS.EDU> Date: Tue, 22-Apr-86 12:58:34 EST Article-I.D.: caip.2003 Posted: Tue Apr 22 12:58:34 1986 Date-Received: Thu, 24-Apr-86 00:04:30 EST Sender: daemon@caip.RUTGERS.EDU Organization: Rutgers Univ., New Brunswick, N.J. Lines: 65 From: Michael Wagner In article <1795@caip.RUTGERS.EDU> LMC-TRANS@GUNTER-ADAM.ARPA writes: > > > >Just got the May issue of Computer Shopper and on page 125 is a short >c program (for the Amiga) that "reads anASCII file out-loud". Has anyone >considered putting such a feature in AmigaTerm, so I could toggle speech on >and speech off in order to listen to my E-mail and only look up to >read it if something interesting came along. I could even listen to it >while it was running in the background and I was doing text >editing or programming in the foreground. I would be very interested in actual experiences of anyone who has tried this. Long ago, when I used to read mail religiously every morning at 300 baud (long ago, when the network was new, one could get through everything every morning in an hour at 300 baud!), I longed for a decent-sounding speech synthesizer so I could listen to NetNews while I drank my morning coffee. When I got my Amiga, I gave some thought to doing this (since I read NetNews with my Amiga). But two problems immediately surface, and I don't quite know how to tackle either one. A) Usenet articles are not straight English. There is a lot of jargon in each and every one. The text-to-phoneme algorithms would have to be augmented or replaced. I don't know what facilities exist within the Amiga to do this, although I expect that one could pre-scan the input, match the jargon words from a look-up table, and provide canned pronunciation for those, and hand the rest off to the translator, glue together the results and give them to the speaker. Not sure how good that would sound. Better yet would be if we could augment the exception tables ourselves. How about it, Amiga? Are the tables user-augmentable? B) Usenet articles generally contain significant visual structure. Tables are sometimes presented. Lines of asterisks and hyphens abound. These contain limited amounts of visual information, but generally no intellectual information. If you were to read an article out to a friend, you would generally skip reading him the lines of asterisks. It isn't clear to me how to get an algorithm to do this, and not skip important information. A similar problem arises with signature lines (talk about no intellectual content!). The reply address is really only of value if you reply. The saying might be of some interest. The graphics don't reproduce in voice at all (if your speech synthesizer algorithm recognizes the octopus in the signature line of the guy from Octopus Systems (or whatever it's called...I forget) and can say "it's signed with that cute octopus" rather than saying "backslash hyphen hyphen space space space oh ...", I *want* to speak to you!). C) [I know I said only two points, but I just thought of this now] The translator and talker, between them, are said to consume over 50% of the available CPU time on an unmodified AMIGA (god knows how one comes up with that sort of number...my AMIGA came without a CPU 'meter' of any form). I'm sure even half an AMIGA's worth of CPU can run a line at 1200 baud, but after you do some semi-intelligent filtering of content-less material out of the articles, I don't think you're going to have much resource left in the AMIGA. You can probably forget about editing in the forground while this runs in the background (unless you like to listen to people who pause a lot between words!). All that aside, I would dearly love to try something that came even *close* to reading NetNews to me. It might make me feel like I'm trying something new and different with my AMIGA. Any takers at solving some of these problems? Michael