Re: IPA speech synthesizer

On Fri, 20 Feb 2009 12:59:30 -0800, Sai Emrys <saizai@...> wrote:

>FWIW: My OP was not to say "could you make it sound perfect", but
>"could you make it sound good enough for Wikipedia use by general
>public on arbitrary IPA-coded pronunciations".
Yes.

>Others, that the sound
>produced, while presumably replicable by a linguistically naïve human
>just from an IPA chart, could not be done by a computer because of
>serial-position combination effects.
I think serial effects are in fact vital to take into account to do what you
want, though.  There are many articulations whose primary acoustic
manifestation is on adjacent sounds: retroflexion, pharyngealisation, etc.
A retroflex [t`] with all the adjacent material snipped off sounds very much
like an alveolar [t] with the same done to it, and anyone who tried to use
such recordings to take away an impression of the difference between the
sounds would probably mistake it to be much smaller than it is.

On top of that, I honestly don't know whether, at a purely acoustic level,
it's easily possible to stitch together recordings of single phones without
getting disruptive gaps or pops or so forth at the phone boundaries.

Alex