[DECtalk] Intelligibility/Listenability criteria

[Don]

Hi Brandon,

On 7/23/2019 1:07 PM, Brandon Tyson wrote:
> These are all really good points. I've been exclusively using Mac and
> iOS lately so have been using Alex extensively. In reading the sample
> text you sent, like "don't desert your friend in the desert") I was
> amazed that Alex was able to differentiate between the contexts and
> read it appropriately.

The synthesizer does this by analyzing the sentence structure and
deducing the appropriate pronunciation from that.  If you just
have "nonsense" groups of words, then all bets are off as to
which the synthesizer will "prefer".

Bob read going black.

Bob read "Going Black".

> No system is going to be 100 percent perfect,
> for instance, in the above sentence Alex says "red" instead of "reed".
> I misspelled read on purpose in order to make sure that any
> synthesizer will speak it properly.

My goal (in the "crippled" synthesizer) is to make sure folks can understand
what is being said without having to interact with the synthesizer repeatedly.
I don't want them to have to ask to have things repeated, spelled, etc.

And, I want to be confident in this even if they haven't interacted with
that synthesizer in years -- as it should only be needed when something is
"broken" or not working as hoped (for example, when the server is overloaded
and can't attend to the needs of yet another user)!