Unique Method To Improve Pronunciation Accuracy Of Homonyms

[Green Hornet]

Greetings everyone!

Although I've been a Nextup customer for many years and have used this forum often, I've never posted before. I'd like to thank those who have contributed in the past for their invaluable and informative information.

Today I thought I'd try to make a contribution too. In my old age I've become increasingly annoyed by the constant mispronunciation of "Homonyms", words that are spelled the same but are pronounced different depending on the sentence structure.

I like to tell my Family and Friends unfamiliar with TTS, that although TTS is not perfect, it is about 98% accurate, but, 100% enjoyable (grin). I've found that almost all of the 2% errors are these annoying Homonyms that always crop up, and makes me cringe during playback. Another portion of these 2% errors is, the peculiar way that a particular voice or speech engine chooses to say a word. Generally the built-in punctuation editor of TA3 can do very little to help with Homonyms.

When TA3 was released, one HUGE new feature I'm loving is the "Watch Mouse" capability (with a few option tweaks), to read highlighted text from within other applications like Microsoft Word. This new feature can quickly and easily, allow me to test known Homonyms to see if that particular voice and sentence structure will mispronounce a word before I create the MP3 audio file.

Excited about this new Watch Mouse tool, one problem remained, locating offending Homonyms or known words a particular voice always has problems saying. After doing some research I did locate a commercial product called "Useful File Utilities" that cost $29.99, but, operates as trialware forever with a brief nag screen timer. After testing this software several times I knew I was onto something very useful in my goal to increase the accuracy of my TA3 audio files.

That being said, I would highly recommend everyone checking out the URL below:

http://www.replsoft.com/index.html

The core software here offers the ability to add plugins to accomplish a wide-variety of functions that I've found very useful. I really like the "Batch Replacer for MS Word" plugin which is needed to accomplish what I describe below. This plugin can be located here:

http://www.replsoft.com/brformsword.html

As I said before, a big problem for TTS are Homonyms, words that are spelled the same but are pronounced differently depending on the context, like, I need to "read" that book, or, I have not "read" that book. To correct the second sentence, you can manually change "read" to "red" to avoid a mispronunciation problem during playback or conversion to audio format.

Using the software listed above, I can use an Excel spreadsheet page to define a word in column "A" and offer a different spelling in column "B". The unique nature of how this software works allow me to not only alter the words automatically, but also define font size, color, or even add highlighting or change other attributes as well. Basically I can scan a 12-point font word document for the word "Minute", and easily change the color to RED and increase the font to BOLD 16-point to help make it standout even more, and also add an alternative spelling in square brackets, like "minute [minit or my-newt]" (without the quotes). Once scanned and altered, I can easily see in my DOC file all the altered BOLD RED words and determine which spelling is needed to correct a mispronunciation error prior to creating an MP3 audio file.

Jim Bretti at NextUp was kind enough to inform me of how to make TA3 ignore any text presented inside square brackets, see directions below:

"To prevent speaking text inside square brackets, add a pronunciation editor entry to a dictionary, setting the Text Matching dropdown to "Regular Expression", and set the Pronounce Using dropdown to "Skip Text". The regular expression you want should look like this:"

\[.*?\]

Once this Regular Expression is added to the pronunciation editor, any words added by the software inside square brackets will NOT be read aloud and can be left as-is in the final document.

Note: Column "A" in the Excel spreadsheet contains the list of words to scan for and alter to what is reflected in column "B". Below are a few more examples of how column "B" might look, minus the larger font size and RED BOLD Color attributes. I always make the first word shown in column "B" as the actual word I was scanning for too, the other square bracketed words are my already pre-calculated spellings to make TA3 pronounce the words as I like if needed.

content [cun-tent or kontent]
dominance [dom-uhnuhns]
Dr. [Drive or Doctor]
Isaac Asimov [Izic Azimov]
live [liv - lyv - liehv - lyive]
project [praw-ject or pro-ject]
record [re-cord or wreckord]
wind [wynd or win'd]
wound [woolnd - woulnd - wownd]

Some words are problematic with several voices, like toothache, so, I can define an immediate change to hyphenate these words to "tooth-ache" for example, yet still define a RED BOLD attribute to the change while also highlighting the background in yellow to better distinguish the type of "automatic" change that occurred.

Using this software and my unique methods of detecting potential problems it locates, I've increased the TTS accuracy from 98% to around 99.5% or higher, the difference in 100% being fully achieved being words that I have never defined in my template and continue to update to improve accuracy even higher and higher in the future.

At this point in time, it becomes a simple process to manually review the MS Word document and quickly test the sentence containing those "RED and BOLD" words using the new "Watch Mouse" feature of TA3 to listen for and correct any potential mistakes uncovered!!!

Microsoft's own native spell checker also underlines in red any ODD spelling of words, which many times leads to being able to correct unfortunate situations the author places within the story. For example; when I converted H.G. Well's "The Invisible Man" recently, one character kept using a guttural English slang to speak, like saying "Ed" instead of "Head" ('E ain't got no 'ed, instead of, He ain't got no head). Using Microsoft Words built in spell check has always be a valuable tool to spot and fix these types of situations before you create your MP3.

Once you use the software I'm recommending, combined with Microsoft Words own built in spellchecker, I'm sure everyone can GREATLY increase their accuracy from 98% to 99.5% or higher as well.

On a side-note, when I began my experimentation's to test this new method of fixing potential problem, I created one master list of Excel words. This list not only included Homonyms, but also included words mispronounced by specific voices. In my humble opinion, I recommend creating a separate Excel spreadsheet that is voice specific, this way you do not see changes to the MS Word document that Heather might make but Paul always says correctly. I have several voices installed, and this mistake of adding voices to the Homonyms spreadsheet, really creates a mess that I need not sort through unnecessarily had I create a separate voice specific spreadsheet too. Yes creating a separate voice list will require scanning a document twice with each spreadsheet, however, it will takes less time to scan twice than to sift through all the potential mistakes one master list could create unnecessarily.

I hope others find my comments and this recommended software useful.

[Bunger Henry]

I'm actually still using TA2, because I can't bring myself to change something I know works. But I wanted to mentioned something I do when I convert a novel to an audiobook. After I have the novel in Microsoft Word, I clear out the custom dictionary in MS Word. Then I spell-check, but I just hold down the button to accept all words into the custom dictionary. After that, I open the custom dictionary in Notepad, and copy all the words into TextAloud. I now have a list of all "unusual" words in TextAloud. I let the program read all the words and type an apostrophe next to all the words that don't sound right. This takes just a few minutes. Then I go in the pronunciation editor and correct all the mispronounced words (typically names, sound effects, and words made up by the author). Now I can convert the novel to audiobook files and the pronunciation problems are very few.