在語音的編碼過程中，除了考慮其編碼效率外，實際運用在編碼系統上尚需加入語音活動檢測(Voice Activity Detection)的前置處理器，以判斷目前的音訊是否為真正語音，若是，則採用編碼系統進一步編碼；若否，則採用簡單的雜訊編碼模擬背景雜訊，這樣不但可以節省傳輸頻寬，更可降低編碼所需要計算量。因此，語音活動檢測在語音編碼上便佔有重要的地位，如何更精準的判斷有聲或無聲狀態，是本論文主要目的。
本研究初步判斷音框為有聲(Voiced)或無聲(Unvoiced)後，再進一步利用頻譜失真(Spectrum Distortion)、越零率(Zero-crossing Rate)、聲道截面積參數和(The Sum of Vocal Tract Area Parameters )、以及音高週期(Pitch)等參數進行第二步的判斷，最後再結合適應性技術，建立一個VAD模型，使其語音活動檢測的判斷能更加精準。
本論文主要以2003年三月份的”大家說英語”教材作為訊練及測試用語料，與GSM和G.729的語音活動檢測方式做比較，其評估後的效果在訓練語料部份整體效果皆較GSM和G.729好；而測試語料部分則僅較GSM好；另外本論文也做了主觀聽覺測試的評估。

Voice activity detector is usually to be the preprocessor of a speech encoder in order to determine whether the incoming signal is a speech segment or not. If it is, a normal speech coder is used to encode the speech segment. If it is not, fewer parameters called silence insertion descriptor (SID) are needed to transmit to the decoder then a comfort noise generator (CNG) is exploited to mimic the background noise. According to the statistics about people’s talking, above 40 % even as higher as 60 % time slice is silence between talk spurts, so lots of bit rates and bandwidth can be saved.
The subject of this thesis is to develop an efficient voice activity detection (VAD) algorithm. There are five speech parameters used to classify the input signal into voiced segments (speech like segments) and unvoiced segments (non-speech like segments) including the segmental energy, the spectral distortion, the zero crossing rate, the fundamental period(pitch), and the sum of the vocal areas. The determination of the proposed VAD model’s parameters and thresholds is based on the steepest descent algorithm. About two-thirds of the teaching material of “Let’s talk in English” in March, 2003 are used as the training database, and the rest is used as the testing database. Finally, the performance of the objective error rate and subjective listening test is studied and compared with the VAD methods of the famous half-rate GSM and G.729 speech coders.

中文摘要 …...i
英文摘要 …..ii
誌謝 ….iv
目次 …...v
表目錄 ...viii
圖目錄 …...x
第一章 緒論……………………………………………………………………1
1.1 簡介………………………………………………………………….1
1.2 過去的研究………………………………………………………….1
1.3 研究動機…………………………………………………………….2
第二章 GSM HR語音編碼器…………………………………………………3
2.1 簡介………………………………………………………………….3
2.2 GSM半率編碼器的語音活動檢測判斷方式……………………...5
2.2.1 適應性濾波器與能量計算(Adaptive filter and energy
computation)…………………………………………………..5
2.2.2 自相關函式的平均值計算(ACF averaging)…………………6
2.2.3 預估值的計算(Predictor values computation)……………….6
2.2.4 頻譜的比較(Spectral comparison)……………………………7
2.2.5 週期的偵測(Periodicity detection)…………………………...8
2.2.6 音調的偵測(Tone detection)………………………………….8
2.2.7 門限值的適應性更新(Threshold adaptation)………………..9
2.2.8 VAD的偵測(VAD detection)………………………………..10
2.2.9 VAD的殘餘增加(VAD hangover addition)…………………12
2.3 GSM半率編碼器的向量和激發線性預測編碼………………….13
2.3.1 線性預估分析(LPC)與內插(Interpolation)係數…………...13
2.3.2 聽覺加權濾波器(Perceptual Weighting Filter)……………..15
2.3.3 長時距預估(Long Term Prediction)………………………...15
2.3.4 固定碼本的搜尋(Codebook Search)………………………..15
2.4 GSM半率編碼器的向量和激發線性預估解碼………………….16
2.5 GSM半率編碼器的非活動語音編/解碼器………………………18
第三章 G.729語音編碼器…………………………………………………….20
3.1 簡介………………………………………………………………...20
3.2 G.729的語音活動檢測判斷方式………………………………….21
3.2.1 有聲/無聲(Voiced/Unvoiced)狀態的初步決定方式……….21
3.2.2 有聲/無聲的平順決策(Voiced/Unvoiced Decision
Smoothing)………………………………………………….25
3.2.3 動態更新四種參數的平均值……………………………….27
3.3 G.729的共軛架構代數碼本激發線性預估編碼器..........................28
3.3.1 前置處理器(Pre-Processing)………………………………..28
3.3.2 線性預估分析(Linear Prediction Analysis)與
量化(Quantization)…………………………………………..29
3.3.3 聽覺加權濾波器(Perceptual Weighting Filter)……………..30
3.3.4 開迴路基頻分析(Open-loop Pitch Analysis)……………….30
3.3.5 適應性碼本搜尋(Adaptive-Codebook Search)……………..31
3.3.6 固定性碼本搜尋(Fixed-Codebook Search)…………………31
3.4 G.729的共軛架構代數碼本激發線性預估解碼器………………...32
3.5 G.729的非活動語音編/解碼器……………………………………32
3.5.1 非連續傳輸(DTX)演算法概述……………………………...33
3.5.2 非活動語音編碼器對無聲嵌入敘述音框的計算與量化….33
第四章 基於最陡坡降演算法之參數加權VAD方法……………………….35
4.1 簡介………………………………………………………………...35
4.2 語料的來源………………………………………………………...36
4.3 VAD所使用的參數及選取各參數的門限值…………………….38
4.3.1 能量參數……………………………………………………..38
4.3.2 越零率參數…………………………………………………..41
4.3.3 聲道截面積參數……………………………………………..43
4.3.4 音高週期參數………………………………………………..45
4.3.5 頻譜失真參數………………………………………………..49
4.3.6 VAD平順(Smoothing)用的參數……………………………51
4.4 VAD判斷參數的權重分數選取…………………………………..52
4.4.1 最陡坡降演算法(Steepest descent algorithm)……………...52
4.4.2 利用最陡坡降演算法搜尋VAD判斷所用的各參數權重分數
值……………………………………………………………..54
4.5 基於最陡坡降演算法之參數加權VAD方法判斷的完整流程…56
第五章 實驗結果……………………………………………………………...60
5.1 評估方式…………………………………………………………...60
5.2 數據分析…………………………………………………………...61
5.2.1 訓練語料的評估結果………………………………………..61
5.2.2 測試語料的評估結果………………………………………..68
5.3 修正VAD判斷方式………………………………………………....73
5.4 修正後VAD判斷的數據結果………………………………………78
第六章 結論與未來展望……...……………………………………………...82
參考文獻 ….83

[1] Antti Pasanen, “Voice Activity Detection in Noise Robust Speech Recognition,” Tampere University of Technology, Master Thesis, April. 2002.
[2] European Digital Cellular Telecommunications System ; Half rate speech; Half rate speech processing functions, ETSI GSM 06.02 (ETS 300-581-1), 1995.
[3] European Digital Cellular Telecommunications System ; Half rate speech; Voice Activity Detection (VAD), ETSI GSM 06.42 (ETS 300-581-6), 1995.
[4] European Digital Cellular Telecommunications System ; Half rate speech; Half rate speech transcoding, ETSI GSM 06.20 (ETS 300-581-2), 1995.
[5] European Digital Cellular Telecommunications System ; Half rate speech; Comfort noise aspects for the half rate speech traffic channel, ETSI GSM 06.22 (ETS 300-581-4), 1995.
[6] General Aspects of Digital Transmission System；Coding of Speech at 8kbits/s Using Conjugate-Structure Algeberaic-Code-Excited Linear-Prediction (CS-ACELP),” ITU-T G.729 Annex B, 1996.
[7] A. Benyassine, E. Shlomot, H. Y. Su, D. Massaloux, C. Lamblin, and J.P. Petit, “ITU recommendation G.729 annex B: A silence compression scheme for use with G.729 optimized for V.70 digital simultaneous voice and data applications,” IEEE Commun. Mag., vol. 35, pp. 64—73, Sept. 1997.
[8] ITU-T G.729, “Coding of Speech at 8kbit/s Using CS-ACELP,” Mar.,1996.
[9] ISO CD 11172-3，Coding of Moving Pictures and Associated Audio for Digital Storage Media at up to about 1.5mbit/s，Part 3：Audio.
[10] 林清池，電腦語音處理與線性預測編碼，台灣：第3波電腦叢書。
[11] R.Tucker, “Voice activity detection using a periodicity measure,” Communications, Speech and Vision, IEE Proceedings I , Volume: 139 Issue: 4 , Aug. 1992, pp.377 -380.
[12] D. K. Freeman, Cosier, G., Southcott, C. B., and Boyd, I, “The voice activity detector for the Pan-European digital cellular mobile telephone service,“ in IEEE Proc. Int. Conf. Acoustic, Speech and Signal Processing, Glasgow, UK, May 1989, pp. 369—372.
[13] F. Beritelli, S. Casale, and A. Cavallaro, “A Robust Voice Activity Detector for Wireless Communications Using Soft Computing,” IEEE Journal on , Volume: 16 Issue: 9 , Dec. 1998.
[14] J.Graybeal, W.Pooch，Simulation：Principles and Methods，台灣：開發圖書有限公司，民國六十九年。
[15] E.Nemer, R.Goubran, S.Mahmoud, “Robust Voice Activity Detection Using Higher-Order Statistics in the LPC Residual Domain,” IEEE Transactions on , Volume: 9 Issue: 3 , March 2001.
[16] European Digital Cellular Telecommunications System(Phase 2); Full rate speech; Transcoding, ETSI GSM 06.10 (ETS 300-580-2), 1995.