# @(#)README.paranoia 1.1 97/04/04 J. Schilling from Monty (xiphmont@mit.edu) README.paranoia Paranoia II: (c) Monty xiphmont@mit.edu Hi there. All CDROM drives are not created equal. You're probably using this patch because yours is a little less equal than others-- or maybe you just keep your CD collection in a box of full of gravel. Jewel cases are for wimps; you know what I'm talking about. This patch adds extra-robust interframe syncronization, code to detect scratches (and hold sync across the scratch) and finally routines to filter out scratches as best possible. These are all handled automatically by fairly modular code. 1) extra interframe syncronization ------------------------------------- Some CD drives can read audio data from an exact starting point to an exact ending point flawlessly; these are rare. A larger number of drives read from only an approximately correct starting point, but do manage to get all the data in-tact from wherever they manage to begin the read. Stock cdda2wav is coded with this in mind. More drives, especially IDE-like and recent ATAPI drives, suffer from framing misalignments within atomic multi-sector read operations. Cdda2wav is also set, by default, to read 75 sectors at a time from ATAPI drives, which Linux's IDE driver breaks into multiple 8 sector reads to conserve kernel memory. Both of these things will break cdda2wav; the symptoms include cracks or pops within the read sample. The "Paranoia" patch will verify the alignment of *every* byte read by cdda2wav. The goal is data integrity, not performance: a minimum 50% speed hit will result from the patch, likely more if your CDROM suffers from the framing bugs the patch corrects. 2) scratch detection and tolerance ------------------------------------ Because overlap syncronization requires matching exact, perfect sample sections, scratched CDs typically cause normal cdda2wav to bail when overlap syncronization is turned on. The second part of the "Paranoia" patch ignores scratches in the read bitstream and syncronizes using the data that can still be "trusted". This step also collects first-pass scratch detection information used by the third section of the patch. Expect performance to drop through the floor on a badly scratched CD; the code can no longer use a simple case of exact matching for speedy correlation. Maintaining solid sync across a scratch is processor bound; later I'll add speedier algorithms for correlation than the brute force method currently used. Scratch detection imposes little additional overhead on a non-scratched CD. 3) scratch repair ----------------------------------------------------- Scratches are an irrevocable loss of data. Still, it's usually possible to reconstruct a sample closer to the original data than the raw hissing, fluttering and crackling that severe scratches in the CD surface produce. Scratch filtering is the 'hard part' of this patch. It needs to do two things; first, find the 'pops' that escaped detection in section 2 of the patch and secondly fill in the gaps. Both are acheived using delta (slew) averaging and forward and backward linear predictive interpolation (with IIR filters) to fill in known gaps as well as look for 'problem values' in stretches of sample known to be scratched. THE SCRATCH DETECTION IS NOT PERFECT. Nor is the repair perfect, although it is easier than detecting scratches reliably. Both are cases of coming arbitrarily close to perfection; Paranoia is designed with light to moderate damage in mind, but will still recover a listenable sample from heavy damage. Paranoia still has quite a bit of room for improvement in the scratch repair code... If cdda2wav + Paranoia is not doing an acceptable job on CDs that you just gotta have, drop me a line; I didn't want to sink months into a project I wasn't certain anyone would ever use :-) Monty