search for: floor_fromdb_lookup

...ecision and the float constants... and there is a large lookup table in the floor1.c ... and I figure out a way to keep the code size and speed, but to improve the precision at this lookup table. (the difference is small, but audible) Here is the modifications in the floor1.c: tatic unsigned long FLOOR_fromdB_LOOKUP[256]={ 0x33E4B434,0x33F390FF,0x3401B286, 0x340A2036,0x34131A1E,0x341CA95A, 0x3426D7A0,0x3431AF44,0x343D3B49, 0x34498767,0x3456A01A,0x346492AE, 0x34736D4B,0x34819F83,0x348A0BF7, 0x3493048E,0x349C9263,0x34A6BF2B, 0x34B19538,0x34BD1F8C,0x34C969DD, 0x34D680A4,0x34E4712D,0x34F3499C, 0x35018C83,0x3509F7B...

Dear All, Floor 1 decoder uses FLOOR_fromdB_LOOKUP table. The purpose of this table is not clear to me. It is also not explicitly explained in the vorbis specification. As i have understood is that it is used to dequantise the "Y" values which are decoded from the bitstream during curve synthesis. Is this correct ? If so why does the ta...

...om it, normally dB_value = 10log(p1/p0), where p0 is the reference value. Supposing p0 is a fixed value, both 10log(p1/p0) => p1/p0 and 10log(p1/p0) => p1 are dB to linear conversions, so, my doubt is: which one of the conversions is done, p1/p0 or p1? Besides, given that the maximum value of FLOOR_fromdB_LOOKUP table is 1, my deduction is that the conversion which is made is 10log(p1/p0) => p1/p0. The alternative would be converting to p1, chosing arbitrarily 1 as the maximum value (in which units?). As I haven't been able to find any reference to this, I tend to believe that the conversion is made...