Cofactor Ora brings people together by facilitating the sharing, systematization, and generation of pronunciation knowledge.
Knowing how words are pronounced is a challenge for language learners and ASR/TTS systems alike. By its very nature, language is a growing, living thing. A traditional (static, fixed-size) word-pronunciation dictionary can never cover all the possible words in a language — neither today, nor tomorrow.
Just look at named entities — streets, celebrities, fictional characters… These change every day, and thus are a practically impossible task for a classical dictionary to identify and maintain.
What if we could use AI to help us do this?
The candidate solution, codenamed Cofactor Ora, is a neural network which works in the cloud to predict pronunciations of English words and named-entities.
Initially trained on common words found in English dictionaries, Ora can be used to infer pronunciations of any number of English words and entity names.
Because it is often the case that only a small group of people knows how to correctly pronounce a particular named-entity (those groups often being local communities in case of street names), Cofactor Ora seeks the involvement of those communities in improving its predictive power.
Through verifying the auto-generated named-entity pronunciations, the users continuously extend the training set used by Cofactor Ora and further enhance its ability to derive accurate pronunciations, making it superior to other G2P algorithms used by current ASR and TTS systems.
Go ahead and start experiencing Cofactor Ora now: