How do I stop SAPI.SpVoice reading "is." as "island"? - text-to-speech

Question

Can you configure the way SAPI.spVoice reads text?
In my situation I am reading the current clipboard using an AutoHotKey script.
The script makes a COM call to SAPI.spVoice passing the text from the clipboard.
;;;;;;;;;;;;;;;;;;;;TTS;;;;;;;;;;;;;;;;;;;;;;
#^!D:: ; Win + Ctrl + D + Alt
ClipSaved := ClipboardAll
clipboard = ; Start off empty to allow ClipWait to detect when the text has arrived.
Send ^c
ClipWait ; Wait for the clipboard to contain text.
ComObjCreate("SAPI.SpVoice").Speak(clipboard)
Clipboard := ClipSaved
ClipSaved = ; Free the memory
return
The problem is.. that SAPI reads some text incorrectly..
For Example:
"Yes it is. Ours is complex." reads "is." as island,
"Yes it is. This is complex." is read correctly.
You can experiment with this by doing the following:
If you are running windows 7.
Press the windows key and type "Change text to speech settings" and pick the option.
In this dialog enter "Yes it is. Ours is complex." in the "Use the following text to preview the voice:" field.
Press "Preview Voice"
Hear it read the "is." as island.
So... My question is...
Is it possible to change/configure the way "Microsoft Anna" reads text so it doesn't make these mistakes?
Is this a bug in the Anna voice only or all voices?
How can I make it read the text the way I want it read?

Answer 1

This is done by SAPI's text normalization code. Unfortunately, this is quite difficult to modify without building a custom voice, which is likely far more work than you probably want to implement.

Answer 2

"Every problem (except the problem of too many levels of indirection) can be solved with another level of indirection."
The SAPI.spVoice object can be passed text (as I was doing) or SSML.
By taking the text to be spoken, then converting it to SSML you gain control over how words are spoken. You have a chance to pre-process the text and replace miss-read words with the specific pronunciation you wish.
For example: "Yes it is. Ours is complex." becomes "Yes it <sub alias="is">is</sub>. Ours is complex."
sub and say_as seem to work. phoneme seem to be ignored.. but I may have something configured wrongly.
Note: If you want XML read aloud, XML escape the text before converting it to SSML, otherwise it will be assumed to be part of the SSML.
So.. in code:
;;;;;;;;;;;;;;;;;;;;TTS;;;;;;;;;;;;;;;;;;;;;;
#^D:: ; Win + Ctrl + D
ClipSaved := ClipboardAll
Clipboard = ; Start off empty to allow ClipWait to detect when the text has arrived.
Send ^c
ClipWait ; Wait for the clipboard to contain text.
FileDelete , c:\tmp\tmp_ahk_tts_clip.txt
FileAppend , %Clipboard% , c:\tmp\tmp_ahk_tts_clip.txt
RunWait, %comspec% /c ""F:\bin\tools\speakit.rb" c:\tmp\tmp_ahk_tts_clip.txt > c:\tmp\tmp_ahk_clip_tts_out.txt" ,,Hide
FileRead, Clipboard, c:\tmp\tmp_ahk_clip_tts_out.txt
ComObjCreate("SAPI.SpVoice").Speak(Clipboard)
Clipboard := ClipSaved
ClipSaved = ; Free the memory
return
and F:\bin\tools\speakit.rb is sometihng like this:
#!/usr/bin/env ruby
substitutions = [
[/[A-Z][A-Z][A-Z][A-Z]+((?=[^A-Za-z])|(?!.))/, lambda{|x|x.downcase}], #All caps becomes word
[/\.exe(?=[^a-z])/i, " executable "],
[/\.txt(?=[^a-z])/i, " text file "],
[/rebranded/, "re-branded"],
[/App(?=[\s\.])/, " application "],
['GUI' , " gooee "],
[/localhost/, "local host"],
[/(?<word>[A-Z][a-z]*)(?=[A-Z ,\.;:\t\/])/, "'\\k<word>' "], # CamelCaseWords should be split by spaces
['\\', '<sub alias="slash">\\</sub>'],
]
require 'cgi'
puts <<-eos
<?xml version="1.0"?>
<speak xmlns="http://www.w3.org/2001/10/synthesis" version="1.0" xml:lang="en-UK">
<voice xml:lang="en-UK">
#{substitutions.reduce(CGI::escapeHTML(ARGF.read)){|o, (r,s)| s.is_a?(Proc) ? o.gsub(r, &s) : o.gsub(r,s) }}
</voice>
</speak>
eos