A Band Extension Technique for G.711 Speech Using Steganography 雖然語音與音訊都是聲音，但是它們卻是有分別的。人類可以發聲的範圍通常稱為語音；而人類可以聽覺的範圍通常稱為音訊。 一般而言，音訊的範圍比語音大。語音之基礎頻寬為4KHz、音訊的基礎頻寬為22.05KHz。 ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB- ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB- ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB- ISLAB-ISLA Author: Naofumi AOKI Source: IEICE Trans. Commun., Vol.E89-B, No.6 June 2006 Reporter: Lin Yu Ying 2007/5/8

Outline Introduction Band extension Transmission of side information with steganography Evaluation Conclusions 2007/5/8

Introduction (International Telecommunication Union) ITU G.711 Such as VoIP (Voice over IP). Encodes speech data into a stream of 8 bit speech samples at an 8kHz sampling rate. Band extension High correlation between the low and high frequency band. 早期發展的編碼技術的發展以「語音編解碼」為主。以簡單的取樣(sampling)與量化(quantization)將語音訊號以數位資料來代表。 由於它的簡單，所以被廣泛地採用。 (Telecommunication Standardization Sector of International Telecommunication Union) ITU-T即將其制訂為標準，稱為ITU-T G.711 2007/5/8

Band Extension Procedure of the band extension technique: 增益gain就是輸入信號經過電路後所放大或縮小的比值 簡單的說，信號經過某電路以後被放大，這個放大的倍數就稱為增益 dB是信號的一種表示它表示的信號非常小以電壓增益來說計算方式是假設增益值是10那它的dB值是20log10=20dB‧ Band-pass(帶通濾波) high-pass(高通濾波) Procedure of the band extension technique: (a) original speech. (b) band-pass filtering. (c) full wave rectification. (d) high-pass filtering. (e) gain adjustment. (f) addition of the low and high frequency bands. 2007/5/8

Procedure of the original technique at (a) sender and (b) receiver Transmission 16kHz 16bit Down sampling G.711 encoding 8kHz 8bit (a) 8kHz 8bit G.711 decoding Up sampling Band extension 16kHz 16bit (b) Procedure of the original technique at (a) sender and (b) receiver 2007/5/8

Transmission of Side Information with Steganography embedding 16kHz 16bit Calculation of gain Down sampling G.711 encoding 8kHz 8bit (a) 旁資訊記錄解碼所須要的相關資料 Side information extraction 8kHz 8bit G.711 decoding Up sampling Band extension 16kHz 16bit (b) Procedure of the proposed technique at (a) sender and (b) receiver 2007/5/8

Transmission of Side Information with Steganography The proposed technique employs steganography to transmit the side information. In each frame, the side information is directly embedded into speech samples by replacing their LSB. Embedding the side information causes some degradation But it is almost negligible since the side information is embedded into the LSB of just 7 speech samples in each frame consisting of 160 speech samples. 最小值位元 2007/5/8

Transmission of Side Information with Steganography Schematic procedure of embedding 4 bit information with the proposed technique (a) Original speech 1 1 (b) sorting 1 1 (c) embedding 1 1 (d) embedded speech 1 1 2007/5/8

Transmission of Side Information with Steganography The average of SNR (Signal-to-Noise Ratio) calculated from ten speech data processed by the proposed technique is 37.903 dB, while the speech data without embedding the side information is 37.908 dB SNR訊號能量與噪音能量比的對數 2007/5/8

Transmission of Side Information with Steganography (b) 若將頻譜圖「立」起來，並用不同的顏色代表頻譜圖的高低，就可以得到頻譜對時間所產生的影像，稱為 Spectrogram Spectrogram 代表了音色隨時間變化的資料 因此有些厲害的人，可以由 Specgrogram 直接看出語音的內容 (a) Original speech " shiro" (b) Band-extended speech " shiro“ with the conventional technique. (c) Band-extended speech " shiro" with the proposed technique. (c) 2007/5/8

Evaluation Ten speech data consisting of 5 male voice (m1~m5) and 5 female voice (f1~f5). The evaluation employed CMOS Comparison Mean Opinion Score 此評估利用CMOS 比較主觀評分結果 此實驗結果證明平均有95%信賴區間 這證明此提出技術可以勝過傳統技術 2007/5/8

Evaluation point quality +3 much better +2 better +1 slightly better about the same -1 slightly worse -2 worse -3 much worse MOS (Mean Opinion Score)為語音編碼品質評估之泛用標準依據。評估方式乃是由一群受過訓練之專業聽眾依照評分以取其平均值， CMOS比較主觀評分結果 Experimental result of the subjective evaluation (proposed vs. conventional technique) Seven-point scale in CMOS 2007/5/8

Evaluation The experimental result shows the 95% confidence intervals of the averages. This shows that the proposed technique may outperform the conventional technique. 2007/5/8

Conclusions The experimental results indicate that the gain adjustment may enhance the conventional technique. Due to the steganography, the proposed technique is able to enhance the speech quality without an increase of the amount of data transmission. 這個實驗結果指出對於增加調節可以提高傳統技術 由於資訊隱藏,這個提議的技術可以不用增加資料傳輸量就能提高語音品質 為了研究這個提議的技術在現實的數位語音通信系統例如VoIP的潛在優勢,更ㄧ步的認證將被考慮之中. 2007/5/8

Thank you so much ! ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB- ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB- ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB-ISLAB- ISLAB-ISLA 2007/5/8

Full wave rectification 2007/5/8

LSB Substitution é 10 00 ù é 2 1 ù S = = ê ú ê ú ë 11 01 û ë 3 û ' 2 4 , é 10 00 ù é 2 1 ù S ' = = ê ú ê ú ë 11 01 û ë 3 û 2 4 2007/5/8

Original LSB Substitution 2 01100111 01100110 01100100 01100101 ú û ù ê ë é = H é 01100110 01100101 ù Z = ê ú ë 01100111 01100100 û 2 2007/5/8

何謂VoIP VoIP(Voice over IP)網路電話，是將語音訊號壓縮成數據資料封包後，在IP網路基礎上傳送的語音服務，透過開放性的網際網路，傳送語音的電信應用服務。利用Internet不僅做到了可即時提供語音服務，更可連接至世界各地，讓使用者可以不需再透過傳統的公眾電話網路(PSTN)進行遠距離電話交談。 back 2007/5/8

VoIP 基本之概念 VoIP 就是將原為類比的聲音訊號以 “ 數據封包 ” ( Data Packet ) 的型式在 IP 數據網路 ( IP Network ) 上做即時傳遞 VoIP 系統就是將原為聲音的類比訊號 數位化後 ( digitized )，透過由網路上各相關通訊協定下，做點對點 ( end-to-end ) 的即時通訊功能。 VoIP 技術可將資料封包在網路上傳遞過程中所發生的失真、迴音及資料遺失做適當修補功能，使其原音重現。 back 2007/5/8

Down sampling Down sampling back 2007/5/8

Up sampling Up sampling back 2007/5/8