Spoken document summarization and retrieval for wireless application

For the purpose of wireless data transformation, spoken document summarization can efficiently reduce the redundant contents. This study presents a voice-activated spoken document summarization and retrieval scheme using text and speech analysis. In this method, prosody, speech recognition confidence, word significance, word trigram and semantic dependency are considered in the summarization score. A dynamic programming algorithm is used to seek the best summarization result. Experimental results indicate that the proposed approach effectively summarizes concise spoken sentences containing important words with semantic dependency.