[MUD-Dev] [TECH] Voice in MO* - Phoneme Decomposition and Reconstruction
Ted L. Chen
tedlchen at yahoo.com
Tue May 21 04:52:49 New Zealand Standard Time 2002
Rudy Neeser Writes:
> For instance, if you want some STT software to operate for people
> from both the south of France and from New York (just pulled those
> places out of a hat), then the software will need to have enough
> data to recognise the same english words being pronounced
> differently by these two different groups of people.
> If your mud is going to have STT software in the client, then
> somehow it will have to recognise the words being pronounced by an
> extremely large and varied group of people (assuming you want an
> international population playing). This is a massive data entry
Quite right. Being a Canadian, I have a tendency to pronounce
certain words differently than my cousins down south :) "Schedule"
being the most common.
From Merriam-Webster Collegiate Dictionary
Canadian also 'she-, British usually 'she-(")dyu(&)l
That was one of the main reasons why I suggest throwing away the
T:text in STT and TTS and replace it with phonemes. At least for
english, we have a finite set of phonemes. It doesn't really matter
which phoneme you choose as long as the computer can recognize it.
So neither the client nor server attempts to understand what was
said, and we rely on the recipient to actually make sense of it.
Whether or not they can from a computer generated voice is something
I'm not sure about myself. I already have a hard time as it is
trying to understand foreign accents in full fidelity.
As for the recent discussions about STT and TTS, I think it's
drifted that way because it's something already familiar to us (and
readily available). I don't mind too much. I still think the
traditional TTS is important to look at design-wise since a feasible
backup system is required in case the player doesn't have a
MUD-Dev mailing list
MUD-Dev at kanga.nu
More information about the MUD-Dev