使用語音輸出入介面之網路即時新聞閱讀系統

Abstract

本論文第一部份製作了一個網路即時新聞閱讀系統，針對網路上即時新聞資料庫(新聞伺服器)作即時的新聞閱讀。它使用隱藏式馬可夫模型語音辨認技術，透過語音命令擷取當日新聞，再利用語音合成系統以語音型式輸出。第二部份為利用改良的SIFT基頻求取法，獲得更為可靠的語音基頻軌跡，結合MLP為基礎的聲調辨認做國語聲調辨認。進一步利用聲調訊息提高系統辨認率，對一個1300詞之獨立詞系統在加入聲調訊息後其辨認率由93.4%提升到94.75%。

In this thesis, two main topics are intensively studied. One is the implementation of a network news reader by using speech I/O interface. It is a real-time news reader by accessing the news server through internet. Using voice command, the user can access different kinds of news article and the content of the article was output through a text-to-speech system. The second topic is to modify the SIFT pitch detection in order to obtain better pitch contour for tone recognition. In an 13000 isolate-word recognition system, the tone recognizer was integrated with a 411 base-syllable recognizer. Performance of the system was examined by simulations using a database containing 960 utterance uttered by 12 speakers. Recognition rates of 93.4% and 94.75% were achieved without/with tone information, respectively.