國立臺灣大學電信工程學研究所李宇旼2006-07-262018-07-052006-07-262018-07-052002-07-31http://ntur.lib.ntu.edu.tw//handle/246246/20250分散式語音辨系統是㆒種架構在無線網路 的語意辨識服務：用戶透過所持有的手持 設備向遠端的伺服器要求辨識的務。由於 在㆗間經過無線通道傳輸資訊，所以在傳 輸過程㆗資料壓縮後產生的量化誤差、無 線通道㆗各種雜訊、傳輸錯誤、頻寬變化 等的影響都會使語音辨識的效能降低，故 分散式語音辨識系統所要克服的問題就是 降低㆖述因素的影響以維持㆒定的辨識 效。本研究是以㆗文大字彙連續語音做為 辨識的對象，希望藉此評估分散式語音辨 識系統的可行性。Distributed speech recognition (DSR) is a client/server approach for speech recognition. In a wireless environment, client refers to the mobile station, while the server is a computer physically connected to the network infrastructure. In DSR, the mobile client performs such processing as feature vector extraction, feature compression, and error control coding. The server, on the other hand, is in charge of more complicated tasks such as speech recognition and understanding, information retrieval, and information processing. DSR is particularly suitable for speech-enabled information retrieval in a wireless environment. However, since the speech feature vectors are transmitted to the server through the wireless channel, feature robustness, tolerance to transmission errors, as well as transmission bandwidth requirements are important design issues for DSR. The purpose of this research is to design practical and robust feature extraction and compression algorithms for DSR that maximize the transmission bandwidth efficiency with minimal loss in information retrieval efficiency.application/pdf339315 bytesapplication/pdfzh-TW國立臺灣大學電信工程學研究所語音辨認分散式語音辨認無線網路Speech recognitiondistributed speech recognitionwireless networksreporthttp://ntur.lib.ntu.edu.tw/bitstream/246246/20250/1/902213E002099.pdf