PHONOLOGICAL RULES FOR ENHANCING ACOUSTIC ENROLLMENT OF UNKNOWN WORDS

Abstract

Phonetic baseforms are the basic recognition units in most speech recognition systems. These baseforms are usually determined by linguists once a vocabulary is chosen and not modified thereafter. However, several applications, such as name dialing, require the user be able to add new words to the vocabulary. These new words are often names, or task-specific jargon, that have user-specific pronunciations. This paper describes a fast method for generating clean, phonetic transcriptions (baseforms) of words using phonological rules on baseforms that have been derived based on acoustic evidence alone [8]. It does not require any prior acoustic representation of the new word, is vocabulary independent, and uses phonological rules in a post processing stage to enhance the quality of the baseforms thus produced. Our experiments demonstrate the high decoding accuracies obtained when baseforms deduced using this approach are incorporated into our speech recognizer. Our experiments also compare the use of acoustic models that are trained on task-specific data with models trained for a general purpose (digit, names and large vocabulary recognition) for the purpose of generating phonetic transcriptions.