Dear All, As i have understood it , the Huffman codebook uses a huffman code present in the bitstream ( lets us say it is decoding a floor 1 Y value using the specifed codebook number ) to walk through the huffman code book tree till it hits a leaf node which is nothing but the entry number. If we consider our floor 1 Y value example , this entry number is directly being used as the Y value. Here is an extract from the tremor code for the same "if((fit_value[j+k]=vorbis_book_decode(books+book,&vd->opb))==-1)" where vorbis_book_decode is returning an entry number. If this is correct the code books constructed for decoding of the floor at the encoder should have as many entries as the maximum value of the floor 1 Y value can take (seems to be 256 ) . In the case of the residue computation the packed entry value decoded is used to generate a residue vecotor ( whose length is equal to the code book dimension ). In this case the number of entries could be more than this . Are these assumptions correct Can someone throw more light on this aspect ? A related question. Is there a limit on the maximum code book entries that could be present in a code book . What does this depend on ? So i have seen a codebook with the maximum number of entries being 6536 ( being used for residue decode i hope ) for the "highnoon1a.ogg" file. warm regards, Sameen -- ___________________________________________________________ Sign-up for Ads Free at Mail.com http://promo.mail.com/adsfreejump.htm -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.xiph.org/pipermail/vorbis/attachments/20050914/1cc9c081/attachment.html
Hi! szerda 14 szeptember 2005 08.30-kor sameen eajaz ezeket a bolcs gondolatokat fogalmazta meg:> code books constructed for decoding of the floor at the encoder should > have as many entries as the maximum value of the floor 1 Y value can take > (seems to be 256 ) .Mostly correct, but there is a multiplier so the codebooks need not have more than 256/multiplier entries. Also for coding efficiency, there are several codebooks, with differing number of entries.> In the case of the residue computation the packed entry value decoded is > used to generate a residue vecotor ( whose length is equal to the code > book dimension ). In this case the number of entries could be more than > this . Are these assumptions correct Can someone throw more light on this > aspect ?I do not understand the question, but the spec probably has the answer.> A related question. Is there a limit on the maximum code book entries > that could be present in a code book . What does this depend on ? So i > have seen a codebook with the maximum number of entries being 6536 ( > being used for residue decode i hope ) for the "highnoon1a.ogg" file.That is a sparse codebook with 81 actual entries. Current encoders don't use more real entries than 1000 (or 500).> warm regards, > > Sameenbye Denes -- --- What kills me, doesn't make me stronger.