Making your machine talk...

[John Bazik]

[[Ed's Note: See also related attempts in Misc digest (v9n391). -bdg]]

In article <513 at brchh104.bnr.ca>, jsb at cs.brown.edu (John Bazik) writes:

|> I was in on that thread.  Someone remembered a long-ago post of an
|> english-to-phoneme program.  I have it now, but I haven't had time to play
|> with it.
|> 
|> The obvious hack is to record all the phonemes and write a backend to
|> string them together and write them to /dev/audio.
|> 
|> If you want the above mentioned program, send me mail.

Um, no, don't send me any more mail!  Yikes!  I've placed the program on
our anonymous ftp archive:

	host:	wilma.cs.brown.edu (128.148.31.66)
	file:	pub/eng_to_phoneme

This past weekend I wrote the audio part.  It's harder than you might
expect.  Parsing the phonemes and slapping audio samples to the speaker
was easy.  Putting together a set of audio samples that sound okay when
concatenated in various combinations is hard.

I wrote it as libspeak.a and one application, scat (speech cat).  I'll
clean up the code and put it on wilma sometime this week.  It'll be
pub/speak.tar.Z sometime on monday, December 10.

If anyone actually gets quality speech out of this stuff - please let me
know.

John Bazik
jsb at cs.brown.edu