[DECtalk] Intelligibility/Listenability criteria
Don
Text_to_Speech at GMX.com
Tue Jul 23 13:10:17 EDT 2019
On 7/23/2019 6:38 AM, Jayson Smith wrote:
> Hi,
>
> A few points here.
> First, I think it's a little of both pronunciation and the voice itself that
> gets on my nerves with ESpeak.
>
> Second, I'd argue that Alex and Alexa do have to contend with unrestrained
> input. If I go into my Alexa web portal and put
> "Sfjsaofhdsahbfiuewfbhifgbfvbiuewqfbirewqfbiwfbiubifdsava" on my shopping list,
> then ask her to read my shopping list, she's going to have to deal with that
> horrible mess of text.
Yes, but what does it end up saying? And, I suspect it says whatever
in a very specific context: "I'm sorry, I don't find anyone who is
selling 'Sfjsaofhdsahbfiuewfbhifgbfvbiuewqfbirewqfbiwfbiubifdsava'".
It doesn't have to contend with trying to apply prosody to
incomplete sentences or a meaningless series of words:
"Bob went snigglepuss" "Here is teh mising peas of the puzl"
I can ensure that all of the messages that I generate for the
synthesizer (either synthesizer!) are grammatically correct. I
can craft them in such a way as to avoid difficult pronunciations.
Or, to exploit known text normalization patterns (e.g., presenting
"2,019" to the synthesizer when I want it to say "two thousand and
nineteen" instead of "twenty nineteen".)
But, I can't guarantee that folks who extend my design will be
as disciplined. Yet, the user will have to contend with the
"input text" chosen by those folks for their extensions!
I don't think it is acceptable (or ethical) to say "that's not
MY problem!"
More information about the Dectalk
mailing list