| 摘要: | 隨著車載電子多媒體系統的迅速發展，使得車內提供的服務無限可能。其中藍芽系統已成為無線通訊技術發展的新領域，這代表著所有的應用將可透過藍芽技術整合功能，讓駕駛人安全無虞地使用行動電話服務，乘坐者便利地享受車上娛樂，而這之中扮演關鍵性的角色將是語音辨識。在此計畫中，我們在首年將針對汽車多媒體環境的發展理念規劃一套多媒體應用語音辨識系統，模擬駕駛員及乘客在汽車上使用多媒體系統的情況。所提出的服務則基於駕駛者在車內環境最常使用的操控模式，包括點播放音樂、撥打電話、上網瀏覽及導航系統等等，透過問答方式的人機互動介面讓操作者感到友善，且本系統中將採用語音合成來模擬人聲以作為回應。我們以關鍵詞萃取為主的辨識技術可提升系統的移植性與擴展性，而階層式架構設計可於汽車環境下增加語音辨識的可靠度。然而車外噪音以及車內環境干擾仍是個挑戰，因此第二年計畫我們將進行車上環境的強健性語音辨識，利用強建語音參數及模型調適等方面的技術來降低測試環境的影響。最後一年計畫中，我們將對系統進一步增建個人化使用的設計，藉由語者辨識技術提供專屬的服務，且再運用語者模型調適技術來強化系統的辨識效能。 ; In this project, we will develop a speech recognition system for multimedia applications to a car. According to the oral command by the car user, the multimedia system will provide the requested service. In this project, some popular services of the multimedia system will be considered. These will include playing the requested music, dialing a phone number, using computer networks, and utilizing a navigation system, etc. To make the communication interface between the user and the system friendly, an interactive dialogue approach will be adopted. The user will input speech signals to the system through a microphone and the system will respond with synthesized human voices pronounced by a speech synthesizer. In the first year, the proposed multimedia system will be implemented by developing a keyword-spotting recognition system for clearer speech signals. A hierarchical structure will be proposed to increase speech recognition reliability in a car environment. However, the noise and interference in a car from its environments will still be a real challenge. Therefore, in the second year, we will dedicate to dealing with robustness speech recognition. Methods based on robust features, model adaptation, and so on, will be considered and tested to reduce the impact of environmental noise. A more robust multimedia system will be implemented in the second year. Finally, in the third year, we will incorporate the system with speaker recognition techniques to establish a more personalized system. It is believed that this will further improve the system performance on recognition. Hence, a prototype of a vehicle multimedia system using a robust keyword-spotting recognition system will be expected and applicable. ; 研究期間 9708 ~ 9807 |
