/* * $Revision: 1.3 $ */ 802.16 IP Telephony Lab會議紀錄 時間:96年12月06日15:00 地點:暨南大學科三館413教室 主講人:王鐘逸 紀錄:林文仁 出席者:吳坤熹老師、吳菖育、張文萍、張兢真、王鐘逸、陳柏州、王嘉裕、顏汎嘉、林文仁、蔡政霖、王筱婷 會議主題:Information Hiding in Real-time VoIP Streams 會議內容: 1.Introduction 2.Proposed Scheme for Speech Hiding *Speech Hiding Space *Compressing Secret Speech *Hiding Secret Speech into Cover Speech 3.Implementing the Scheme 4.Experimental Result 5.Reference 問題討論: Q1. 0.041 應該唸成 "zero point zero four one", G.711 則唸成 "g dot seven eleven".(老師建議) Q2. 這個方法小於1ms適用於 realtime 的 VoIP application ,以此做為你的 conclusion.(老師建議) Q3. 投影片第22頁。You said you can hide a secret speech into the cover speech, but I don't understand that this seems to be a larger file and how can you hide a larger voice file into a smaller file?(老師提問) A3. We choose the cover speech whose length is 48 seconds. We will repeat it and send it to the receiver.(鐘逸回答) Q4. 投影片的內容都是句子,可是為什麼都沒有句號?(文萍提問) A4. 如果有句號就都要句號,所以就沒用句號。(鐘逸回答) A4.因為他有些不是句子,所以他擔心有些有,有些沒有,如果統統都沒有句號這是可以的。(老師回答) Q5. 投影片第2頁的 Outline 的 introduction 沒有列出項目。(文萍建議) A5. 這裡會有不同觀點,大家還記得李家同老師演講那一次,事實上他建議大家這一頁,除非你給大家 一個 overview ,說明主要是用什麼樣的貢獻,用什麼樣的方法來達到什麼樣的效果,不然這些名詞, 當你還沒往下講之前,這些名詞都對所有的聽眾都是無意義的。所以這頁當所有的名詞都還沒定義時, 花太多時間在上面,底下的人不見得會知道。除非用High-level的overview,不然的話,此頁一閃而逝, 或直接跳過此頁都是可以的。此頁可以帶給聽眾的資訊是很少的。 不過文萍點出的問題是後面的 introduction 太冗長了,兩三頁就夠了,有些可改成 Motivation 或是 Component、Codec,這樣會更清楚。(老師回答) Q6. 投影片第9頁,H263少了一個點。應該是 H.263。(文萍提問) Q7. 前面鐘逸唸 "g dot seven hundreds eleven" 這是最完整也點冗長的唸法,一般常聽到人家唸"g seven eleven"(G.711), "h two sixty three"(H.263),沒有唸 "hundreds"。 我們通常取小數點後兩位比較容易唸。 ms 唸 "millisecond" dB 唸 "decibel" 一般還是看習慣來唸。(老師建議) Q8. 投影片第16頁,160沒寫單位,突然看到會感到怪怪的。(柏州提問) A8. 我會將單位加入這頁。 Q9. 投影片第16頁,第二項這個句子不順,若是我會將它改寫成 In Linphone, each G.711 μ-law packet will contain 160 samples. 這樣較簡短。 在第一次將你的投影片完成之後,第二天再回過來看一次投影片,找出有哪個句子不順,並加以修改。 這樣文章的Quality就會越來越好。(老師建議) Q10. formula(1)後面沒有講到。(柏州提問) A10. 此formula是本論文唯一的formula。論文有formula的話,被接受的程度較高。(老師回答) Q11. 投影片第13頁,這兩個方法為什麼會使用第2個比較好,有什麼差異?(菖育提問) A11. Figure 1. means one speech packet, there are 160 samples. We choose hiding samples by interval 1. Figure 2. means one hiding sample. We choose least r bits as part of hiding space.(鐘逸回答) Not two methods.(筱婷回答) Some people may get confused, Chang-Yu thought there are two different methods, and the second one is better than the first one. So you choose the second one. In fact, there is only one method presented in this paper. The upper is the first step. The lower is the second step. So maybe the better way for your figure to present the cover speech, is like this. 1 2 3 4 5 6 7 8 bit +-+-+-+-+-+-+-+-+ 0 | | | | | | | | | +-+-+-+-+-+-+-+-+ 1 | | | | | | | | | <-- +-+-+-+-+-+-+-+-+ 2 | | | | | | | | | +-+-+-+-+-+-+-+-+ 3 | | | | | | | | | <-- +-+-+-+-+-+-+-+-+ ... | ... | +-+-+-+-+-+-+-+-+ 159 | | | | | | | | | <-- +-+-+-+-+-+-+-+-+ packet Each packet has eight bits. The way you presented causes the confusion.(老師回答) Q12. "Would you repeat again?",這一句是很典型的句子,文法沒錯,但是通常美國人不會這樣講, 雖然用 "would you" 已經是很客氣了,但是美國人還是習慣你加"please","please" is a magic word. 所以完整的句子是 "Sorry, could you please repeat your question again?" (老師建議) Q13. Do you recommand that the size of the hiding space is not too big to affect the original cover speech? (兢真提問) A13. Hiding space 在一個 packet 所占比率多大才不會影響聲音的品質。(老師解釋) 一個 packet 如果用 3bits 來藏,聲音品質會不太好,藏 1bit 就可以接受, 所以比率是10/160=6.25%。(鐘逸回答) Q14. 投影片第14頁,表格忘了加 "Table 1",後面有提到"Table 1"。(兢真提問) Q15. 投影片第24頁,S->S'是什麼意思?(嘉裕提問) A15. This is the time of Speex compressing and decompressing.(鐘逸回答) Q16. 投影片第5頁,句子敘述有的字有粗體,有的沒。(政霖提問) A16. 粗體是我要強調那一段。(鐘逸回答) 強調 Key word、專有名詞,很少是強調一整行的,除非真的整行都要。(老師回答) Q17. How many experiments do you make to get a set of data that satisfies the scenario you design?(筱婷提問) A17. We play this cover speech whose length is 48 seconds. We play this cover speech eight times. So we get 384 seconds.(鐘逸回答) Q18. What dose the D stand for?(老師提問) A18. It's the secret speech which has been decompressed.(鐘逸回答) Q19. I don't know how to get the formula 1.(汎嘉提問) A19. In the formula, it can calculate how many bits every hiding sample could contribute. That means we select odd indexes in cover speech and these samples are called hiding samples. So for anyone hiding sample we must to caculate the bits are. It means how many bits which every hiding sample need to contribute. So we get the formula.(鐘逸回答) Q20. 不管是用中文或是用英文發問前,試著將完整的句子寫下來。(老師建議) Q21. Some words are pronounced incorrectly, for example, method, experiment. (文仁建議) 散會: 結束時間:17:36