EXPANDING AN EFFECTIVE VOCABULARY OF A SPEECH RECOGNITION SYSTEM

摘要

The present invention provides technology for creating and using segmentation word model, with increased effective dimensions active vocabulary speech recognition system. The active vocabulary indicates what all words and this speech recognition system of phrase were able to recognize that. Each word can be by combining acoustic model. In this way, this speech recognition system of the combination of acoustic model of active vocabulary expression can be compared, to identify the voice of the acoustic model best match user of those user speech. Effective dimensions active vocabulary can increase segmentation word to composition component or segment (for example, prefix, suffix, separator infix and root), and be individual entry active vocabulary including each component. Thus, for example, word list and their plural form (such as, " book, book, the cooking are cooked; hook, hook, appearance seems ') can be indicated on active vocabulary using word (such as; " book, cook, hook, and look up and down ') and entry represent suffix make the words plural number (such as, "+s, here ' the s ' of '+' before indicate that '+s+ ' suffix). For large list, word, word suffix and also ignore associated entry, which can reduce entry, need to indicate that word list is quite big.