kosma%human-torch@stc.lockheed.com (Monty Kosma) (06/01/90)
Here's a question along the lines of the recent discussions of speech synthesis on the amiga. Can somebody give me a brief rundown or pointer to a reference on how it works? Does the amiga actually have a speech synthesis chip, or is it done through audio samples of phonemes, or what? And the real question, how good **can** it be with the current **hardware** limitations? monty kosma@alan.decnet.lockheed.com
peterk@cbmger.UUCP (Peter Kittel GERMANY) (06/08/90)
In article <20704@snow-white.udel.EDU> kosma%human-torch@stc.lockheed.com (Monty Kosma) writes: > >it works? Does the amiga actually have a speech synthesis chip, or is >it done through audio samples of phonemes, or what? ^^^^^^^^^^^^^^^^^^^^^^^^^ Yes, this way. And there is very tricky software to bind these phonemes together and make fluent speech of them. What is to be said about quality: The max. sample rate of the Amiga beyond 18 kHz is far enough for speech (regard the bandwidth of a telephone with 3 kHz!), only limitation is in the quality of the digitized phonemes. Frankly speaking, they are such a pure Texan style American (just for a German ear) that they are nearly useless for other languages than American. I believe even Britons would need different phonemes. All this was done by an external software company and sold to Commodore. They even had offers to prepare it for other languages, but the charge was so expensive that noone took it. -- Best regards, Dr. Peter Kittel E-Mail to Commodore Frankfurt, Germany rutgers!cbmvax!cbmbsw!cbmger!peterk
kosma%human-torch@stc.lockheed.com (Monty Kosma) (06/09/90)
> >it works? Does the amiga actually have a speech synthesis chip, or is >it done through audio samples of phonemes, or what? ^^^^^^^^^^^^^^^^^^^^^^^^^ Yes, this way. And there is very tricky software to bind these phonemes together and make fluent speech of them. What is to be said about quality: The max. sample rate of the Amiga beyond 18 kHz is far enough for speech (regard the bandwidth of a telephone with 3 kHz!), only limitation is in the quality of the digitized phonemes. Frankly speaking, they are such a pure Texan style American (just for a German ear) that they are nearly useless for other languages than American. I believe even Britons would need different phonemes. All this was done by an external software company and sold to Commodore. They even had offers to prepare it for other languages, but the charge was so expensive that noone took it. -- Best regards, Dr. Peter Kittel E-Mail to Commodore Frankfurt, Germany rutgers!cbmvax!cbmbsw!cbmger!peterk hmm, so let's say I (or somebody) wanted to improve on this, or just fool around a bit. Would having different/better samples help? Or does the "binding together" software need work? What I guess I'm asking is what is the weak link that makes the voice so poor. monty
bard@jessica.stanford.edu (David Hopper) (06/09/90)
In article <21526@snow-white.udel.EDU> kosma%human-torch@stc.lockheed.com (Monty Kosma) writes: > > > >it works? Does the amiga actually have a speech synthesis chip, or is > >it done through audio samples of phonemes, or what? > ^^^^^^^^^^^^^^^^^^^^^^^^^ > Yes, this way. And there is very tricky software to bind these phonemes > together and make fluent speech of them. What is to be said about quality: [...] > -- > Best regards, Dr. Peter Kittel E-Mail to > Commodore Frankfurt, Germany rutgers!cbmvax!cbmbsw!cbmger!peterk > >hmm, so let's say I (or somebody) wanted to improve on this, or just fool >around a bit. Would having different/better samples help? Or does the >"binding together" software need work? What I guess I'm asking is what >is the weak link that makes the voice so poor. This has indeed been done. I have a program called 'Talk' that was written by one Jon L. Sherling on 6/18/88. To include the documentation (I have the source but I am hardly a programmer, yet): (Note: this is to be read by Jon's 'Talk' program, so forgive the atrocious phonetics) "This is the TALK program written by John Sherling on June eighteenth, nuynteen eighty eight. It is an attempt to reeplace the Amiga speech synthesis with my own voice. Digitized phownemes created with fewchersound are red in to the program and used in conjunction with the Amiga translaytor luybrary. The phownemes are in standerd fewchersound format and can be reeplaced in order to improve the voice quality or to change the voice entuyerly." It is a valiant attempt and a fascinating-sounding voice, although I'm not sure if it is more understandible. Certainly, it is more humanlike. I'd post the source, but it's 11416 bytes, and I figure this is long enough already. > >monty Dave Hopper | /// Yesterday, CS. | My favorite icebreaker: | /// Today, Anthro/History. | "If you were really my bard@jessica. | \\\/// | friend, you'd kill me Stanford.EDU | \XX/ Tomorrow... bleeding ulcers. | now."
LEEK@QUCDN.QueensU.CA (06/09/90)
I wouldn't mind using extra memory for something that sounds a little better - how about a sexy female voice for my Amiga :) (Research shown that pilot are more alert to female voice than a male voice from the on-board instrumentations) ) Hmmm. I don't need research to tell me that... :) Seriously I think now that most Amigas have enough memory for better samples of phonemes and may be samples for different inflection instead od computed ones. This should improve on the quality of speech synthesizer if memory is available and more and more applications would use this ignored feature of the Amiga. K. C. Lee