An Automatic Lexicon Collection Method Using Wisdom of Crowds for Speech Interfaces of Vehicle Information Services

The authors are currently developing a speech interface that enables users to obtain information without the need to look or touch the display of mobile terminals such as smartphones. The proposed interface uses speech recognition and synthesis technology, and has potential applications in vehicle information services. To accurately read aloud texts and correctly recognize users' speech, the authors developed a function that automatically collects current terms and proper nouns based on wisdom of crowds in the Internet. More specifically, it is a function that automatically extracts orthography and description of Japanese syllabary characters (furigana) for each of the words in Wikipedia, estimates part-of-speech information, and registers them into a word dictionary of the system. The accuracy with which text was read was improved by a morphological analysis using the word dictionary collected by this function, together with a basic word dictionary prepared beforehand. In a reading accuracy evaluation, misreading decreased by about 40% in texts such as news sites and sightseeing information, which are frequently used in vehicle information services.