Pronunciation Discrepancy: Main vs Basic Editing Windows

[speaktext]

Why is that a word that is mispronounced in the main TextAloud window will often be pronounced perfectly correctly when you enter it with the same spelling in the Basic Pronunciation Editing window (or, sometimes, in the main TA window all by itself)?

George

[kremesch]

This is just a guess. But if I recall correctly, the Pronunciation editor of TA pronounces words without taking into consideration any other edits that you might have on the word.

Is it possible that you have the word(s) that you're having trouble with set in another editor (i.e. the vendor's pronunciation editor?) or through a RegEx expression?

It might help to know what word(s) you're having trouble with and from what vendor as well.

[speaktext]

Thanks, kremesch, but, no, there's no other editor installed/involved here.

There have been a number of words that have demonstrated this discrepancy between the main TA window and the Basic Editing window, but the word I came across today was "shove". The source was a news story with the phrase "the shove they need to reform". From experimenting, I know that when the following conditions are met, the word is pronounced *correctly* in the main window:

1. "shove", with no other words in the main window
2. "the shove", with no other words in the main window
3. "shove they", with no other words in the main window

When the phrase "the shove they" is in the main window, however, the word "shove" sounds like "shibe" or "shive".

The voice I'm using here is ScanSoft Tom22. AT&T Mike16, by the way, does fine with "shove", but pronounces the word "they" like "ay".

George

[RicochetRita]

When the phrase "the shove they" is in the main window, however, the word "shove" sounds like "shibe" or "shive".

Try changing the text to:

the shove: they
or
the :shove they

I've found that doing this sometimes forces the engine to re-evaluate a troublesome word without its neighbors (although it adds a pause between the words).

Also, if a problem word is followed by a comma, replacing the comma with a semicolon can have a similar correcting effect, but with a less noticeable pause.

[speaktext]

Thanks, RicochetRita. Putting the colon *after* the word "shove"--either in the main window or in the Basic Editing Pronunciation field--does force a correct pronunciation.

I still find this behavior somewhat curious, though. I recognize that the context for a word is an essential part of speech-to-text software (speech recognition), but I don't see why context should affect text-to-speech software like TextAloud (except where it involves a part of speech issue, e.g., "produce" as a verb versus "produce" as a noun).

Can anyone shed any light on why this problem exists for text-to-speech software? Is this kind of problem actually due more to the particular limitations of the *voice* software (e.g., ScanSoft Tom22) than the TextAloud software? As I've indicated, different software voices stumble over different words/pronunciations, but it's not clear to me that this definitely indicates that the problem is attributable, either wholely or in part, to the voice app.

Thanks.

George

[speaktext]

Any input from the TextAloud development team regarding the issues raised here would be appreciated.

George

[kdwhite]

In general it is all about context. Some of the voices have pretty complex algorigthms that try to understand a word based on what is around it. This can mean a word alone might be pronounced perfectly, but when in a sentence with other words nearby, the voice will try to pronounce it a little differently. While sometimes this is great, it can make it a real challenge to get edits right.

[speaktext]

Ken,
Thanks for your reply. I'd appreciate it if you would clarify a point that you made.

In a previous post, I said "I don't see why context should affect text-to-speech software like TextAloud (except where it involves a part of speech issue, e.g., "produce" as a verb versus "produce" as a noun)". You've now said that, "a word alone might be pronounced perfectly, but when in a sentence with other words nearby, the voice will try to pronounce it a little differently".

Unlike what I said on this point, you do not qualify your statement with a reference to a word involving different parts of speech that normally affect pronunciation. So, would it be correct to infer that you are saying that a software voice takes context into account when determining pronunciation all the time, for each and every word, regardless of whether the word is one (e.g., "produce") that is normally (i.e., by humans) pronounced two different ways, according to its part of speech status?

And, if so, why? Why wouldn't words whose pronunciation never varies be coded to indicate that, so that the voice software would ignore the context when determining pronunciation?

Thank you.

George

[kdwhite]

George, the answer is, we don't really know. Because we don't make the voices but people way smarter than us do (we hope they are anyway), it is often a mystery what is under the hood. So I don't really know the answers as to why/how.

[speaktext]

Thanks for the clarification, Ken. You have, by the way, implicitly answered one of my questions: these pronunciation anomalies are caused by the voice apps, not TextAloud.

George

[Gelo30]

yah its cool Ken, I also learn a lot from you. . .

[Joeguy]

Up to now, I have been building up a dictionary homonyms where they are classified incorrectly. So I would have an entry bow for the weapon, and an entry boow for the action (and its noun counterpart). It is somewhat tedious, and I can't always find the perfect pronunciation (even with the use of online dictionary phonetics).

Given that both pronunciations are present in the speech engine, since most of the words are pronounced correctly in one context or another. I was wondering who does the word classification based on the context? Is it TextAloud or the Speech Engine? And in either case, is there a way to manually override incorrectly identified contexts?

[kdwhite]

It is at the speech engine level and not much of a way to change it.