Hi,
I had a 1 gb CF card, which was full of samples, doodles and sketches that I gathered over time, work in progress material. One day I turn on my SP-404 and get that blue screen of death equivalent - flashing FNT signal. Part of me just accepted it and valued the experience of done work, but the other part just couldn't let all those ideas go. I've searched this forum for possible solutions, but couldn't find anything, that could resolve the problem, so I started experimenting.
The filesystem of CF looked ok through computer, so i tried copying to the other card, the whole ROLAND folder. No luck. Then I tried to copy the folder of one bank (e.g.) BKUP3 to the working card with other samples on it, BKUP3 was not empty on that card, therefore had to be overwritten. This worked, I could load those lost samples. I then tried overwriting other BKUP* folders, on both empty and full folders - everything worked. But then I tried copying SP404 folder and the new card flashed FNT, so I looked inside and compared the working and non-working. The problem was 2 files were missing PWDCHECK.SPD, WVCINFO0.SPD. After putting those files in from a working card (the content of the card did not matter) the card started working.

All in all:
-check the content of folders of both working and non-working cards
-find what's missing and copy those files from a working card
-if everything is in place - copy the "non-sample" files from a working card. those can be found intuitively. Or copy "sample" files to the working card

p.s. there were also a lot of unsuccessful guessings of what's wrong, data recovery softwares etc, and i tried a lot of different combinations of copying files back and forth, some of them worked some of them did not. the tips i wrote - worked for me at the end. sorry for the long post, but I had to spill it out somewhere, it was a big win for me =]

UPD.: I've just noticed, the sample length is different then of those original samples, like they are cut off. It's not the "Mark" button. Will have to investigate further...

After analyzing the contents of WVCINFO0.SPD, which turns out to be some sort of indexing file for the 404, I was able to figure out some patterns. Similar as in the thread about 404sx hack viewtopic.php?f=3&t=6831&hilit=hack

The file consists of 4416 symbols. Divided by the number of avalable pads (12), through all the banks (C-J, we geta line of 46 symbols per pad:
000000000000B5100000B51000000101007F023A00023A

00000000 0000B510 0000B510 - those parts are sample start, end, length. If we take hex number B510 and convert it to decimal, we get 46352. If it's divided by 44.1, the result is 1051... which equals to the length of a sample in ms, i.e. 1051 ms, or 1.051 s.

00 00 01 01 00 - those numbers correspond to lo-fi, stereo, gate, loop and reverse in the same order. 00 - the red light on the button turned off, 01 - button is lit up. e.g. 01 in second position would mean the sample is in stereo.

7F - that's for sample volume, 00-7f = 0-127
further it gets a bit shady for me
023A - i think it's sample BPM range, like when the 2 knob is turned when TIME/BPM is pressed and the range goes from 50 to 100 to 200
00 - not sure what that is, but i have seen it having 00, 01, 02 values
023A - the BPM of the sample.

For the sampler to recognize SMP*.SMD files, it is necessary to have correspondent lines lines in the file. I've created the blank file, by formatting the card on the sp and then just entered sample end and length (same numbers) and the stereo flag (there are 2 files for stereo samples SMP*L.SMD and SMP*R.SMD, * being the number of the sample). The BPM values were left default. They gained new values when the bpm was corrected on the sampler. When the length of the sample in the index file is shorter than the actual sample - it gets cut off on playback. If it's longer - no problem, the samples gets long which a tail of silence at the end. It can then be truncated on the sampler.

I was trying to figure out how to calculate the length of the samples by their size, but did not find any dependancies between sample time and SMP*.SMD file size. If anyone knows anything about this - please let me know.