Pronounciation Server

This is an experimental pronounciation server; it is a part of the Sphinx Knowledge Tools.

This server is guaranteed to return a pronunciation. Initially it looks up your word in the Carnegie Mellon 100,000 word dictionary (a limited form of affix processing is used to increase the hit rate). If a pronunciation can't be found there, the server attempts to create a pronunciation using rules from text-to-speech synthesis. In the latter case, the qualilty of the pronunciation is not guaranteed, as it would not have been checked by a human expert. Since the synthesis component is tuned for English orthography, expect problems to arise in the interpretation of words with a non-English origin. Also, bear in mind that this module does not attempt to figure out how certain orthographic forms ought to be rendered in spoken form. For example, "St." might variously be spoken as "street" or "saint". Digit sequences will be rendered as such, with no attempt to turn them into numbers or currency amounts. Such transformations are in the realm of text normalization.

Enter words to be pronounced, one per line:

Please bear in mind the following usage notes:

• The system can deal with numbers (as digit strings) and can deal to some extent with punctuation, no serious attempt is made to condition the input text. The user should do this on their own, as a pre-processing of the input.
• The output from this server has stress marks stripped out. If you need these, you should use the Carnegie Mellon Pronouncing Dictionary.
• Multiple pronunciations are notated as WORD(n), where n refers to a particular variant. Variants are guaranteed to be unique, however n is not guaranteed to increment sequentially.
• The output may be saved to a file and used as any other recognition dictionary. Be sure to save the file in "source" form, to preserve the delimiter between word and pronunciation (which is a TAB character, octal 011; otherwise tabs found in your input will be replaced by spaces).
• At this time, orthographic words that exceed a certain length are (silently) ignored. The limit is around 35 chars (about the length of six average words). Non-alphanumerics are also ignored.