pilgrim@daimi.aau.dk (Jakob G}rdsted) (04/29/91)
This posting is about the softwareimplemented
speech capability of the Amiga. Well, I'm not
satisfied with it ! The problem seems to be
the phonemes. They are allright for talking
american(english?), but not for talking
danish. I trust in a lot of replies on this,
since germans must feel just as let down, since
it can't say u umlaut either(well can it?)
I've looked in the manuals, and there is a
mention of a standard, that the amiga confines
to. Why does this standard only support phonemes
used in english ? Isn't there a standard with
more sounds, like some decent r's, a proper
"u umlaut(y for danes)" etc?
So what I want to know is why the standard chosen
was chosen, and what one can do to get u umlaut,
and decent r's, etc.
--
From the notorious
Jakob Gaardsted, Computer Science Department
Bed og arbejd ! University of Aarhus, Jylland (!)
(Pray and work!) AMIGA! | I'd rather play Moria.jsibley@magnus.acs.ohio-state.edu (James Sibley) (04/29/91)
In article <1991Apr28.195319.3987@daimi.aau.dk> pilgrim@daimi.aau.dk (Jakob G}rdsted) writes: >This posting is about the softwareimplemented >speech capability of the Amiga. Well, I'm not >satisfied with it ! The problem seems to be >the phonemes. They are allright for talking >american(english?), but not for talking >danish. I trust in a lot of replies on this, >since germans must feel just as let down, since >it can't say u umlaut either(well can it?) >and decent r's, etc. >-- >From the notorious > Jakob Gaardsted, Computer Science Department >Bed og arbejd ! University of Aarhus, Jylland (!) >(Pray and work!) AMIGA! | I'd rather play Moria. On a related (?) note, I've heard that there is a PD program out there somewhere which allows you to use digitized samples as phonemes to the "say" command. If anyone has any info on this product, I'd appreciate relaying it my way.. Thanks.. -- James Sibley Nous Sommes Du Soleil Seeking the truth about lemon curry. We Are Of The Sun jsibley@magnus.ircc.ohio-state.edu We Can See Amiga: the only true computer. -YES-
peterk@cbmger.UUCP (Peter Kittel GERMANY) (04/29/91)
In article <1991Apr28.195319.3987@daimi.aau.dk> pilgrim@daimi.aau.dk (Jakob G}rdsted) writes: >This posting is about the softwareimplemented >speech capability of the Amiga. Well, I'm not >satisfied with it ! The problem seems to be >the phonemes. They are allright for talking >american(english?), but not for talking >danish. I trust in a lot of replies on this, >since germans must feel just as let down, since >it can't say u umlaut either(well can it?) You're completely right. We all know this and there also has been a discussion inside Commodore about this. It has not yet led to any real result, but you can be sure, we talk about it *now*. There has even one Danish colleague moved to West Chester, but I'm afraid he will bother with some other aspects, or will he? >I've looked in the manuals, and there is a >mention of a standard, that the amiga confines >to. Why does this standard only support phonemes >used in english ? Isn't there a standard with >more sounds, like some decent r's, a proper ^^^^^^^^^^ >"u umlaut(y for danes)" etc? ^^^^^^^^ I love you :-) >So what I want to know is why ... Well, besides other aspects, it's also a $$$ (or was it even more $'s?) question... And, to make things more difficult, WOULD IT REALLY PAY? How many applications would use it? I see that many current English/American applications don't use speach, though they could. And in CDTV, one will go straight digitized speech. -- Best regards, Dr. Peter Kittel // E-Mail to \\ Only my personal opinions... Commodore Frankfurt, Germany \X/ {uunet|pyramid|rutgers}!cbmvax!cbmger!peterk
UH2@psuvm.psu.edu (Lee Sailer) (04/30/91)
There are two somewhat separate issues here. One is how
typed text is automagically translated to phonemes, and the
second concerns how the phonemes get converted to speech.
For the best quality speech, you can send strings of phonemes
synthesiser without any translation. The little SAY utility is
just a little demo program. It doesn't show all of the Amiga's
capabilities very well. For example, even Amiga Basic will translate
ordinary text into strings of phonemes, which you could save
in a file. Then, see how those sound, and edit them directly
if you feel the need.
As I recall, those phoneme files are in something called the "ARPABET",
which is a "no special characters" version of the widely used IPA, or
International Phonetic Alphabet. The IPA is supposed to be able to make
all (nearly all??) sounds from any human language, but you have to
know the IPA "spelling." (In this case, you need the ARPABET spelling.
I think they used to be in the BASIC manual, of all places.)
Probably, the routines that translate text "How are you?" into
the ARPABET "/HAW AAR YEW" is probably tuned to American English.
The synthesis routines shouldn't be, but might be by ommission if the
code was developed in the US.
One last point. Believe me now and hear me later. The Amiga does
not do such a good job on English, either. I'd be surprised if
it could do any worse on Danish. Ja? Oui? Da? Ano? Maalob? Yes?
leebard@jessica.stanford.edu (David Hopper) (04/30/91)
Before summer vacation last year, I promised I'd put the 'talk' program
up on ab20. I've had horrendous luck uploading to my ultrix account,
but I think I've figured out what's wrong. I'll archive the 'talk'
binaries and phonemes and put them up soon. I promise.
'Talk' was written by one John Sherling, and allows you to use your own
phonemes in conjunction with the translator.device. It sounds pretty
cool, but has about the same functionality as the built-in speech (if
not a bit less, because you have to use the 'talk' binary, and the
loading time is a bit long for all the phonemes). It's worth a listen,
though, because it is so much different from the built-in device.
A big plus, though; if I recall, the source is distributable. Hack away.
Another plea: isn't there any way to rewrite the translator.device to
support foreign phonemes?
Peace,
Dave Hopper | /// Anthro Creep | Academic Info Resources, Stanford
|__ /// . . | Macincrap/UNIX Consultant
bard@jessica. |\\\/// Ia! Ia! | -- Just remember: love is life, and
Stanford.EDU | \XX/ Shub-Niggurath! | hate is living death. :Black Sabbathjap@convex.cl.msu.edu (Joe Porkka) (04/30/91)
UH2@psuvm.psu.edu (Lee Sailer) writes: >There are two somewhat separate issues here. One is how >typed text is automagically translated to phonemes, and the >second concerns how the phonemes get converted to speech. The single biggest reason, IMHO, is because the Amiga was designed, built, and software was written, in the USA. {of course it helps that each human laguage needs its own custom set phonemes} So, if you were on a tight budget (time and money), and you were writing a speach synthysizer, wouldn't you make it speak your native language? This of course is no excuse for the continuation of the english-centrism, especially considering howmany Amiga sales are in non-english speaking countries.
GELSON%SBU.UFRGS.ANRS.BR@uicvm.uic.edu (Gelson Dias Santos) (04/30/91)
Lee Sailer <UH2@psuvm.psu.edu> writes: >As I recall, those phoneme files are in something called the "ARPABET", >which is a "no special characters" version of the widely used IPA, or >International Phonetic Alphabet. The IPA is supposed to be able to make >all (nearly all??) sounds from any human language, but you have to >know the IPA "spelling." (In this case, you need the ARPABET spelling. >I think they used to be in the BASIC manual, of all places.) I'm trying to make the amiga speak in portuguese, and I have the same problem. I can find the correct IPA phonemes for portuguese words in good dictionaries, but how can I convert them to ARPABET ? Is there any kind of "conversion table" somewhere? >One last point. Believe me now and hear me later. The Amiga does >not do such a good job on English, either. I'd be surprised if >it could do any worse on Danish. Ja? Oui? Da? Ano? Maalob? Yes? It speak better than I :-) * Gelson Dias Santos * Bitnet/Internet: GELSON@SBU.UFRGS.ANRS.BR * * Porto Alegre - RS * Sorry, english is not my native language! * * BRAZIL * (Alguem fala portugues?) *
peterk@cbmger.UUCP (Peter Kittel GERMANY) (04/30/91)
In article <91119.165012UH2@psuvm.psu.edu> UH2@psuvm.psu.edu (Lee Sailer) writes: > >Probably, the routines that translate text "How are you?" into >the ARPABET "/HAW AAR YEW" is probably tuned to American English. Yes, they are, definitely. (Though I don't whether you only mean the difference between American and British English.) You need totally different routines to convert other languages. Though again, for many other languages it should be easier to write such routines than for American, because other languages are more phonetic, so they have a more direct correlation between spelling and pronouncing than English, without so many exceptions. >The synthesis routines shouldn't be, but might be by ommission if the >code was developed in the US. And they are too tuned tightly to American. The phonemes are all spoken with the tongue stuffed way back in the throat, whereas e.g. German in contrast is pronounced with the tongue more stretched flat. (I'm no linguist, but this way it appears to me when speaking both.) So to pronounce other languages correctly with *their* normal accent, also this stuff should normally be adapted. -- Best regards, Dr. Peter Kittel // E-Mail to \\ Only my personal opinions... Commodore Frankfurt, Germany \X/ {uunet|pyramid|rutgers}!cbmvax!cbmger!peterk
peterk@cbmger.UUCP (Peter Kittel GERMANY) (04/30/91)
In article <1991Apr29.232924.23182@leland.Stanford.EDU> bard@jessica.stanford.edu (David Hopper) writes: > >Another plea: isn't there any way to rewrite the translator.device to >support foreign phonemes? Well, as it's totally different for other languages, it's more than some additions to existing code, it's a complete rewrite, for every language. Some attempts have already been done, by an Italian company (Cloanto, commercial), and by me for German (well, *I* understand it, but other men at least need some time to tune in and get accustomed to the horrible sound, they don't understand a word for the first minutes...). This is a pure prototype stadium, so I'm hesitating to publish it. -- Best regards, Dr. Peter Kittel // E-Mail to \\ Only my personal opinions... Commodore Frankfurt, Germany \X/ {uunet|pyramid|rutgers}!cbmvax!cbmger!peterk
monta_l@dist.dist.unige.it (Luciano Montanaro e Marco Gualdi) (05/06/91)
The phonemes are too few! I also want phonemes to support foreig(my!) languages. I can't get properly spelled r's, double letters, gn(spanish n~) gl(spanish ll) etc. Then there should be different translator.libray version for the different languages ... (well, the important thing is a larger base of phonemes.) Ciao a tutti! Luciano Montanaro monta_l@dist.dist.unige.it
masaru@media-lab.media.mit.edu.MEDIA.MIT.EDU (Masaru Sugai) (05/07/91)
In article <1991May6.111801.1641@dist.unige.it> monta_l@dist.dist.unige.it (Luciano Montanaro e Marco Gualdi) writes: >The phonemes are too few! >I also want phonemes to support foreig(my!) languages. >I can't get properly spelled r's, double letters, gn(spanish n~) gl(spanish ll) >etc. >Then there should be different translator.libray version for the different >languages ... > >(well, the important thing is a larger base of phonemes.) > > Ciao a tutti! > Luciano Montanaro > monta_l@dist.dist.unige.it Folks in Europa are much happier than I am! A few months ago I tried to have Amiga speak in Japanese, but I gave it up. As for the phonem I had little problem as Japanese has rather poor language in phonetics, so I could manage to get my favorite ones combining lots of phonemes. The difficulty comes from that fact Japanese makes use of changes in pitch a lot to distinguish words which have the same spelling. For example,'ha-shi' means both bridge and chopstick in writing, but we can tell the difference as bridge changes its pitch from high to low, and chopstick from low to high. That is the reason English spoken by most Japanese sounds very monotonistic. I heard Chinese has a more complex articulation system (four voice), and in general Asian languages have similar characteristics in common. I don't demand Amiga engineer to be a linguistist :), but I would be pleased if narrator device allowed me to have a lower level control, especially stress/pitch attenuation in sub-phonemes level. BTW, this is a good chance to appeal my idea in the world. What about putting your ARPABET name in your signatures ? I always find difficulty to imagine correst pronunciation of people on the other side of the Atlantic. ARPABET is really pain in the neck, but far better than nothing... -- SUW5GAH5IH5 MAH5SAH5WLAXUH5 (Oh! I must have forgotten Japanese :) -- -- Masaru Sugai:Use disclaimer. CIS 72050,2141:NeXT + A3000 = money-eater NEC Corporation:sugai@ccs.mt.nec.co.jp DORMANT:hardwired logic,machine language MIT R.Affiliate:masaru@media-lab.media.mit.edu: "Silicon on Sapphire" by CLASH
dannie@coplex.uucp (Dannie Gregoire) (05/08/91)
While we're on the subject.... Can anyone tell me where the phonemes for the voice synthesizer are stored and how they are stored?? I have always wondered if it was possible to resample the phonemes to make my amiga speak better (and with my voice ;-) Any ideas?? +-------------------------------------------------------------------------+ | Dannie J. Gregoire \\\\//// dannie@coplex | | Copper Electronics Inc. ////\\\\ !uunet!coplex!dannie | +-------------------------------------------------------------------------+
peterk@cbmger.UUCP (Peter Kittel GERMANY) (05/08/91)
In article <1991May7.211150.16383@coplex.uucp> dannie@coplex.uucp (Dannie Gregoire) writes: > >Can anyone tell me where the phonemes for the voice synthesizer are stored >and how they are stored?? > >I have always wondered if it was possible to resample the phonemes to make >my amiga speak better (and with my voice ;-) Any ideas?? No, obviously they are not stored in a sampled form. It must be a set of numerical parameters for every phoneme, I think. (Guess: I had the same idea and searched long in the narrator for a single phoneme, but found only one sine curve.) -- Best regards, Dr. Peter Kittel // E-Mail to \\ Only my personal opinions... Commodore Frankfurt, Germany \X/ {uunet|pyramid|rutgers}!cbmvax!cbmger!peterk