網路電話的語音品質的問題是自網路電話開始發展之後就一直被提出來研究的課題，然而現今的分散式網路架構基於頻寬共享的特性，只有提供“盡力而為“的網路服務，對於資料延遲較敏感應用程式，如網路電話，其語音品質會受封包延遲、遺失、抖動等網路損傷因子所影響。本論文研究目的在於研究並發展出可以動態調整網路電話語音品質之整合系統。首先，整合主動式(PESQ)與被動式(E-Model)的語音量測方式，可以即時監控網路語音品質的變化。然後，利用田口實驗設計法做參數設計，選擇靜音偵測(VAD)、不同的聲音檔、語音編碼器、封包延遲時間、封包遺失率、封包重複率、回音消除緩衝區(EC-Tail)、封包編碼長度(ptime)作為實驗的控制因子；實驗中，語音品質以望大特性作為目標特性，並選用 直交表進行實驗，用變異數分析(ANOVA)找出顯著因子及百分比貢獻度，再以確認實驗來驗證實驗的再現性。另外將影響語音品質的顯著因子結合倒傳遞類神經網路來建構語音品質預測功能，並利用田口實驗設計法找出類神經網路的最佳學習參數，最後將類神經網路的預測功能整合至網路電話軟體PJSUA，依據網路環境來預測語音品質，然後透過語音編解碼器的速率控制來進行動態調整語音品質。

The voice quality issue of Voice over Internet Protocol(VoIP) has been studied since the development of VoIP. However, the distributive network structure is based on bandwidth sharing, thus, can only provide network service to its limit. To programs sensitive to data delay, such as Internet Protocol Phone(IP Phone), the voice quality would be affected by packet delay, loss, vibration, and other damaging factors. The purpose of this study is to development and implementation of an integrated system for dynamically adjusting VoIP voice quality. First, active Perceptusl Evaluation of Speech Quality(PESQ) and passive E-Model voice measuring methods are integrated to monitor the network voice quality in real-time. Then Taguchi method is used for parameter design, and Voice Active Dection(VAD), different voice files, voice encoders, packet delay time, packet loss rate, packet repeat rate, Echo Canneller tail(EC-tail), and PTime are selected as controlled factors in the experiment. In the experiment, the target characteristic of voice quality is the larger the better, and orthogonal array is used for experiment. Then, Analysis of Variance(ANOVA) is used to find significant factors and percentage contributions, then confirmation experiment is conducted to verify the reproducibility. The significant factors that affect voice quality are combined with backpropagation neural network to construct the voice quality predicting function. Taguchi method is used to find the optimal learning parameters for the neural network. Last but not least, the predicting function of the neural network is integrated into to the IP phone software PJSUA, in order to predict the voice quality based on the network environment. Then, voice encoder rate is controlled to dynamically adjust the voice quality.

[1] ITU-T Recommendation G.107, “The E-Model, a computational model for use in transmission planning”, ITU-T, 2000.
[2] ITU-T Recommendation P.862, “Perceptual evaluation of speech quality (PESQ), an objective method speech quality assessment of narrow-band telephone networks and speech codecs”, ITU-T, 2001.
[3] P. Galiotos, T. Dagiuklas and D. Arkadianos, “QoS Management for an enhanced VoIP platform using R-factor and Network Load Estimation Functionality”, High Speed Networks and Multimedia Communications 5th IEEE International Conference, pp. 305-314, 2002.
[4] L. Ding and R. A. Goubran, “Speech quality prediction in VoIP using the extended E-Model”, Global Telecommunications Conference, vol 7, pp. 3974-3978, 2003.
[5] L. Carvalho, E. Mota, R. Aguiar, A. F. Lima, J. N. de Souza, “An E-Model Implementation for speech quality evaluation in VoIP Systems”, Computers and Communications, ISCC 2005. Proceedings. 10th IEEE Symposium on , pp .933-938, 2005.
[6] N. O. Johannesson, “The ETSI computation model: a tool for transmission planning of Telephone Networks”, IEEE Commun. Mag., pp. 70¬79, 1997.
[7] J. W. Seo, S. J. Woo, and K. S. Bae, “A study on the application of an AMR speech codec to VoIP”, In Proceedings of IEEE ICASSP, pp. 1373-1376, 2001.
[8] L. Roychoudhuri, E.Al Shaer, “Real-time audio quality evaluation for adaptive multimedia protocols”, IEEE Management of Multimedia Networks and Services(MMNS), Barcelona, Spain, 2005.
[9] Z. Qiao, L. Sun, N. Heilemann and E. Ifeachor, “A new method for VoIP quality of service control use combined adaptive sender rate and priority marking”, IEEE International Conference on Communications, Paris, France, Volume 3, pp.1473 – 1477, 2004.
[10] S. Chimmanee, W. Komwut, R. Suwan, “Hybrid neuro-fuzzy based adaptive load balabcing for delay-sensitive internet application”, Journal of Intelligent and Fuzzy Syatem, vol 16, pp .79-93, 2005.
[11] Meddahi, H. Afifi, D. Zeghlache, “Packet-E-Model:E-Model for wireless VoIP quality evaluation”, IEEE International Symposlum on personal, Indoor and Mobile Radio Communication Proceedings, pp.2421-2425, 2003.
[12] 湯鴻沼, “第七號共通信號系統概論”, 全華通書, 1999.
[13] Hersent, J. P. Petit, D. Gurle, “Beyond VOIP protocols”, Wiley, 2005.
[14] K. Wallace, “Voice over IP First Step”, Cisco Press, 2006.
[15] 陳宏宇, “VoIP網路電話技術”, 文魁資訊, 2005.
[16] 逢愛君, “VoIP先進技術原理與應用”, 自強工業基金會, 2004.
[17] 賈文康, “SIP會談起始協議操典”, 文魁資訊, 2006.
[18] J. Anderson, “Communications test equipment”, Agilent Technologies, 2002.
[19] D. Hardman, “Noise and voice quality in VoIP environments”, Agilent Technologies, 2002.
[20] J. Anderson, “Methods for measuring perceptual speech quality”, Agilent Technologies, 2001.
[21] H. Zhang, L. Xie, J. Byun, and C. Shim, “Packet loss burstiness and enhancement to the E-Model”, Software ngineering, Artificial Intelligence, Networking and Parallel/Distributed Computing, 2005 and First ACIS International Workshop on Self-Assembling Wireless Networks. SNPD/SAWN 2005. Sixth International Conference, Page(s):214 – 219, 2005.
[22] ETSI EG 201 050 V1.2.2, “Speech processing, transmission and quality aspects (STQ); Overall Transmission Plan Aspects for Telephony in a Private Network”, ETSI, 2002.
[23] 吳富強, “田口品質工程”, 全威圖書, 2002.
[24] 李輝煌, “田口方法品質設計的原理與實務”, 高立圖書, 2003.
[25] 蘇朝墩, “品質工程”, 中華民國品質學會, 2005.
[26] L.Sun, “Speech quality prediction for voice over internet protocol networks”, Ph.D thesis, University of Plymouth, 2004.
[27] 張裴章, 張麗秋, “類神經網路”, 東華書局, 2005.
[28] 王進德, 蕭大全, “類神經網路與模糊控制理論入門”, 全華圖書, 2005.
[29] 葉怡成, “類神經網路模式應用與實作”, 儒林圖書, 2003.
[30] 林毓達, “網路電話VoIP技術與測試驗證實作班”, 工業技術研究院, 2005.
[31] 王榮宗, “網路電話(IP Phone)產品設計”, 自強工業基金會, 2006.