2013 | OriginalPaper | Buchkapitel
Constructing Language Models for Spoken Dialogue Systems from Keyword Set
verfasst von : Kazunori Komatani, Shojiro Mori, Satoshi Sato
Erschienen in: Contemporary Challenges and Solutions in Applied Artificial Intelligence
Verlag: Springer International Publishing
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by (Link öffnet in neuem Fenster)
Spoken dialogue systems (SDSs) need language models (LMs) for automatic speech recognizers (ASRs) for each domain. This is because domain-specific words such as proper nouns differ in domains, and they must be recognized correctly to accomplish the task.We propose a method to construct a class N-gram LM only from a set of domain-specific words (i.e., words in target relational database for retrieval). This problem setting corresponds to a situation where we construct a new spoken dialogue system; i.e., there is no sufficient corpus available in the target domain.We use a similar-domain corpus and assign class labels to it using machine learning. Because no sufficient training data are available, we create an initial training corpus by string matching and then use it as training data. The experimental results showed that our approach is promising: ASR accuracy for domain-specific words improved.