Page 1 of 1

Voice Manipulation and Pronunciation Changes

PostPosted: Thu Dec 07, 2006 2:19 pm
by MultiPort
Cepstral offers several ways to manipulate word and phrase pronunciation.

For instance:

1. First, the voice is sensitive to punctuation and word order. You might try rearranging your sentence, or verb selection, or adding commas or such to troubleshoot the issue.

2. Cepstral offers a lexicon.txt file where users can make global changes word pronunciation. Instructions are on our FAQ at: For example, if you had a weather related application and wanted always to say "wind" in context and avoid ever confusing the wrist watch context of "wind". You would add the following to a new line in your lexicon.txt file:
    wind 0 w ih1 n d
3. SSML - We support the SSML control tags that might help you adjust pauses or add emphasis to clarify word or phrase meaning. Our FAQ also covers SSML briefly at:

4. You can also mix the two (Cepstral's lexicon + SSML) such that you send phonetic strings dynamically to the engine wrapped in the PHONEME tag. Examples:
    <phoneme ph="g ay1 k ow0">Geico</phoneme>
    <phoneme ph="l i1 d">lead</phoneme> ... as in The lead singer is Mick Jagger
    <phoneme ph="w ih1 n d">wind</phoneme> in The wind outside today is 5 MPH
5. Similarly, if you favor other phone sets, you can use them too via the PHONEME tag. For example, SAMPA, a popular European phone set could be used as in the following example:
    <phoneme alphabet='sampa' ph='gaIko'>Geico</phoneme>
    *Note SAMPA phones does not have spaces, and Cepstral's do.

6. Lastly, Cepstral offers professional services. We can "tune" your prompts at a lower level than what SSML can. If needed, we can even have the original human speaker add your prompts to the TTS voice database. Check with sales for this option.


-Cepstral Sales

TextAloud offers some pronunciation alteration as well.

PostPosted: Mon Jan 15, 2007 11:42 pm
by TaoPhoenix
This kind of option may tie into the lexicon behind the scenes; I'm not sure.

I found it vital to get "OK" to quit being "Oklahoma". There are a few other horrid ones as well that escape me at the moment. I think these originate with Microsoft's fundamental engines.

Anyone willing to share a lexicon.txt file?

PostPosted: Sun Jan 28, 2007 12:10 pm
by Shaggyman
I have just started with a Cepstral voice I have licensed with the intention of producing audiobooks to contribute to the Internet Archive for free distribution. I want to correct the pronunciation of as many words as possible. Does anyone out there have a lexicon.txt file they are willing to share? I will eventually post mine, but I have just started it and it contains only a few words.

Leslie Evans

lexicon.txt error messages

PostPosted: Sun Jan 28, 2007 12:15 pm
by Shaggyman
I have just started to work with adding words to lexicon.txt, and as suggested have been using the run swift command to preview how words sound. it took a while to realize when I tried many variant vowels trying to get the right sound but nothing was changing that an error message was being flashed on the screen for a few seconds saying that it was ignoring my attempted phonemes as invalid. There was no apparent reason why any possible vowel sound couldn't be placed after a consonent, so I have no way of knowing why my various choices were ruled invalid. Are these rules written down anywhere?

Also, the ony way I could view this list of rejected words was to hit the Print Screen button as it flashed by and dump the screen into PhotoShop. Surely if there is going to be an error message it should stay on the screen long enough to read it.

PostPosted: Fri May 18, 2007 10:22 am
by Alex
Hi Shaggy,

The general format is:

word 0 w er1 d
wordtwo 0 w er1 d t uw1

Some things that can prevent a userlex entry from being valid are:
    * Capitalization of the entry: (Word 0 w er1 d) - or - (word 0 w eR1 D)
    * Leaving out the zero after the word: (word w er1 d)
    * Not having stress on a vowel phoneme: (word 0 w er d)

Hope this helps,


PostPosted: Thu Dec 20, 2007 12:09 pm
by Tom55
I'm trying out Cepstral voice for the first time and I really seem to like it!

Not working

PostPosted: Thu Mar 11, 2010 12:55 pm
by mthakershi
I am trying to instruct my system to say simple words in proper pronunciation.

2 examples
fine 0 f ah1 y n
pressure 0 pr esh1 er0

Swift throws error and ignores entries.
What are the rules? Can someone point to a document. I did see basic cepstral support page that has some information but there is no way one can build words from that help.

Thank you.


PostPosted: Tue Apr 13, 2010 1:25 pm
by mthakershi
Why is no one replying to my question? Please help.

PostPosted: Sat Apr 24, 2010 6:08 pm
by AdamW
The simple reason that no-one is replying is because the answer to your question is posted as the first posting to this thread. Please read the first post of this sticky thread as it give the link to an online page describing the format, location and rules of the user lexicon.