Breaking the Language Barrier


SpeechCorrect is a completely unique process for teaching the speech recognition program Voice Pro 12 to distinguish between homophones.

Homophones are words that have the same pronunciation but are spelled differently depending on the context. They are one of the big, well-known problems for anyone working with dictation. The different ways of spelling, and thus the meanings, can, in fact, only be determined by the context. For example, the spelling differences in “world peace” and “a piece of cake” are easy to spot, although they sound exactly the same.

The comprehensive extension of the speech modules in Voice Pro 12 can pick up on the majority of such homophones; after all, the Linguatec corpus used for its development comprises no less than 700 million word forms! The context analysis is based on a statistical process which runs in the background when you are dictating and calculates which spelling should most likely be used. The prediction applies so-called bi- and trigram statistics. If a dictated bigram or trigram is contained in the system, or if it has already been used by the speaker, then this is decisive for the correct choice of words.

Still, there are situations where the speech models and their statistical probability will not suffice. Take the name “Smith”, for example: it is one of the most common names in the English-speaking world, and exists in any number of forms with alternative spelling. Do you know someone called (or is your name) Joanne Smith, David Smythe or Claire Smyth? The speech model cannot tell which form is the correct one on its own. However, with a practical and intuitive form, users of Voice Pro 12 can establish that “Joanne” should always be followed by “Smith”. This way, you can also clearly distinguish between, for example, the famous fashion designer “Paul Smith” and “Smyths Toy Store”.

Even challenging phrases like “The chili from Chile is excellent, but the climate there is rather chilly”, are no match for SpeechCorrect. To Linguatec, it was essential that the user specific rules be simple and do not require previous knowledge of linguistics.



Contact | Imprint | General Terms and Conditions
Copyright © 2018 Linguatec. All rights reserved.