臺灣博碩士論文加值系統

改進型多頻帶激勵 (Improved Multi-Band Excitation, IMBE) 語音編碼系統是以多頻帶激勵(MBE)編碼模式為基礎發展而來。由於它對於激勵源的處理特色，使其可以用較低的位元率合成良好聲音品質。具體而言，IMBE編碼系統是將每一個音框在頻域上以三個諧波(Harmonic)組成一個子頻帶(Sub-band)，再分別對每一個子頻帶判別其為濁音子頻帶或清音子頻帶。因此，有別以往傳統語音編碼器在同一音框中只能判定成全濁音或者是全清音，IMBE語音編碼器可以在同一音框中同時容許濁音與清音的特性共存，使合成語音品質更貼近自然語音。
本篇論文的研究重點是探討IMBE語音編碼器隨著子頻帶總數上限(原始設定為12)減少，隨之而來的位元率降低及可能的聲音品質劣化。我們的實驗結果證實，在固定的4000Hz系統頻寬內，基頻高的女聲品質較不受減少子頻帶總數影響，例如基頻170Hz的女聲只需要8個子頻帶。反之，只要基頻低於121Hz的男聲就需要12個以上的子頻帶，因此如果將子頻帶總數上限降至12以下，就容易對基頻較低的男聲品質產生影響。本研究也證實，如果願意接受整體音質略降，在語音內容維持合格辨識度的要求下，可以透過適度減少子頻帶總數以降低位元率。

The improved multi-band excitation (IMBE) vocoder is developed from the multi-band excitation (MBE) speech model. It can provide good speech quality with low bit rate, primarily due to its special form of treating excitation signals. Specifically, for each speech frame, through spectral analysis, the IMBE vocoder divides the whole spectrum into sub-bands, each consisting three harmonics. Then each sub-band is discriminated as voiced or unvoiced. By doing this, the IMBE vocoder allows a speech frame having voiced and unvoiced sound at the same time, i.e., the nature of mixed sound. And this is why the IMBE vocoder can reconstruct synthetic speech with more nature quality.
In this thesis, for the IMBE vocoder, with decreasing upper limited number of sub-bands (the original setting is 12), we investigated the corresponding bit-rate reduction and the possible quality degradation of synthesized speech. Our experiments proved that, with the 4000 Hz system bandwidth, high-pitched female voice will be affected less due to the decreasing of sub-bands. For example, only 8 sub-bands are required for a voice with 170 Hz fundamental frequency. On the contrary, more than 12 sub-bands are required for voice with fundamental frequency lower than 121 Hz. Our experiments also showed that, within acceptable speech intelligibility, the bit-rate reduction can be achieved if a certain amount of speech quality degradation is tolerable.

誌謝 i
摘要 iii
ABSTRACT v
目錄 vii
表目錄 ix
圖目錄 xi
第一章 緒論 1
1.1研究動機 1
1.2章節架構 2
第二章語音型成/特性/模型原理 3
2.1語音型成 3
2.2語音特性 4
2.3語音模型 4
第三章IMBE語音編碼器 5
3.1 IMBE語音編碼器簡介 5
3.2 IMBE語音編碼器與一般傳統語音編碼器的差異 5
3.3 IMBE語音編碼端 6
3.3.1音高估算 7
3.3.2音高細估 10
3.3.3濁音和清音的判斷 11
3.3.4頻譜振幅 12
3.4 IMBE語音解碼端 13
第四章 調整Sub-band 降低位元率之探討 14
4.1降低位元率概念 14
4.2降低位元率步驟 16
第五章 實驗結果 18
第六章 結論 31
第七章 參考文獻 33

[1] J. P. Campbell, “The New 4800 bps Voice Coding Standard,” in Proc. Mil. Speech Tech. 89, Washington D.C., pp. 64-70,Nov. 1989.
[2] D. W. Griffin and J. S. Lim, “Multiband Excitation Vocoder,” IEEE Transactions on ASSP, Vol. 36, No. 8, pp. 1223-1235, Aug. 1988.
[3] Telecommunication Industry Association , “T.I.A-102.BABA_Vocoder
Description,” TIA standard, 2002.
[4] 程強、崔慧娟、唐昆(2003)。低速率語音壓縮編碼對語音識別系統的影響。計算機應用研究2003/09期:22-28。
[5] 侯紅芳、張太鎰(2009)。多帶激勵語音編碼的實現。現代電子技術2009/07期。第294期:163-166。
[6] 林威延(2005)。基於聽覺特性之LPC編碼之研究。電機工程研究所，碩士論文。台北:中原大學/ 電機工程學系。
[7] 語音辨識-技術與應用,[Online]. Available: http://agms.tku.edu.tw/pc/web_main/ppt/ppt_01.ppt