陳佳新：碩士論文與成果

碩士論文與成果

中文題目：網路電話語音品質改進之研究
英文題目：Speech-Quality Improving for Internet Telephony
中文摘要：為了舒緩封包遺失對於語音品質的影響，本論文分別對網路電話的傳送端與接收端程式，研究有效的補償方法。本論文提出的傳送端補償方法，在不惡化傳送失敗率的情況之下，改變封包傳出失敗的分佈，這對接收端的波形重建處理非常有幫助。實驗結果顯示，若使用我們的方法，連續傳出失敗的發生次數中，個數為1與2所佔之比例，將可由26.01%提高至73.62%，而導入的平均傳出延遲則為2.12個封包處理時間。本論文提出的接收端補償方法，首先會去偵測遺失封包的前一個與後一個封包內的基週長度，接著決定重建出的基週在遺失封包內需要佈放的個數與個別長度，最後內差出各個基週的波形，此方法我們稱之為時間比例式基週波形內差法（TPBPWI）。主觀的聽測評估實驗顯示，我們的方法比雙邊基週波形複製法（DSPWR）、改良式波形相似性疊加法（Modified WSOLA）、相似波形取代法（PM）與過去封包重複法（Rep）等波形重建方法都好。在窄頻寬的撥接網路實驗中，主觀的聽測評估結果也顯示，使用TPBPWI方法可以提升語音品質。
ABSTRACT：In this thesis, we study to develop efficient methods to mitigate the effects of packet losses in internet telephony. The methods proposed include a transmitter-based PLC (Packet Loss Concealment) method using a packet dropper and a receiver-based PLC method. The proposed transmitter-based PLC method changes the distribution of packet transmission failures. That is very helpful for the receiver to reconstruct the waveform of a lost packet, and the rate of transmission failure is not degrated. When compared with the original method in the ratio of burst packet-loss length of 1 and 2, our method is shown to have higher ratio, i.e. 73.62% vs. 26.01%. Also, the transmission delay induced is only 2.12 packet times. The proposed receiver-based PLC method detects the pitch lengths of the waveform in the two sides, determines the number of pitch cycles and pitch lenghts to be placed into the lost packet, and interpolates the pitch waveforms from the two sides to synthesize pitch waveforms. This method is called Time-Proportion Based Pitch-Waveform Interpolation (TPBPWI). When compared with other waveform reconstruction methods, e.g. Double Sided Pitch Waveform Replication (DSPWR), Modified Waveform Similarity OverLap Add (Modified WSOLA), Waveform Substitution based on Pattern Matching (PM) and Packet Repetition (Rep), our method is shown, in subjective perception tests, to be able to obtain better speech quality than the methods mentioned. In addition, our method, TPBPWI, has been integrated into WinRTP program and tested with a dial-up network of narrow bandwidth. The result of perception-based evaluation shows that our method can obtain significant speech-quality improvement.
目錄：緒論／封包遺失補償方法之回顧／傳送端補償方法／接收端補償方法／模擬實驗與評估／補償方法與WinRTP之整合／結論
口試現場：中華民國九十四年七月二十五日下午兩點三十五分／中華民國九十四年七月二十五日下午三點十六分
其他照片：中華民國九十三年九月三日ROCLING-2004研討會於翡翠灣福華飯店（我是左邊數來第二位）
DEMO音檔：原始音檔／以白努力模型模擬遺失30%封包時的音檔／使用Packet Repetition補償後的音檔／使用TPBPWI補償後的音檔
DEMO波形：使用Packet Repetition封包遺失補償方法／使用TPBPWI封包遺失補償方法
實驗室：國立台灣科技大學／資訊工程研究所／語音暨聲訊處理實驗室
指導教授：古鴻炎博士
研究生：陳佳新