Now you can! With the help of Lord Nightmare, we can make your dreams come true! Download the sidtestdisk 4.0C1.d64, and make me a Hi-Fi recording of the output of the program.
Modify your C64. Take out the SID chip and locate the third pin from top right (number 26, I think). It is the AIN (audio in) pin, and it sucks in most of the system noise, especially the display whirr. We need to ensure it is grounded or floating. The easiest fix is to make it floating: bend it slightly off, so it doesn't enter the socket on chip insertion. If you are very adventurous you could also try to improve the SID input voltage with additional capacitors and coils. It will help a little, but this modification is what counts.
Each run of the program makes 2048 / 8 = 256 measurements of about 5.6 seconds each. Therefore it takes about 24,5 minutes to run the program once.
You need to run the program twice, once selecting 1=highpass and once 3=lowpass test. I do not need or use the bandpass filter result. You should get two samplings with very quiet filtered sounds, with the filter cutoff increased each time.
Do try to ensure that the output is high quality. Use 48 kHz sampling rate, 16-bit mono. There is no need to go for 96 kHz or 24 bit. There isn't that much detail in the recording.
Do not post-process the audio in any way. At most compress it with a lossless codec such as FLAC.
Each sample is delimited by a loud snap, courtesy of the filter, which is used by my splitter program to cut the samples into pieces for analysis.
The sound is going to be very quiet, so you can probably turn volume pretty high up, if you can. Try to tune it such that the loud snaps hardclip just barely. Use the highpass test for checking the sound level, as it is loudest at the beginning. Do not adjust level during and between the actual recordings. This would break the analysis.
The distortion is a thorny problem. I rather start from the nondistorted sound and look into adding it, than start from the distorted and try to work the rest of it out. I have got pretty good model for the distortion anyway, so I can add it quite easily. The nondistorted FC curve is the most fundamental artifact I can get.
Make them available on the web, and ping me at ircnet using nick alankila or mail me at firstname.lastname@example.org. Alternatively, you can run the scripts yourself. fc-curves -directory contains the current scripts and pile of sampled chips.
The output is type 1 -style definition. Fit to type 3 definitions so far is manual. Please use type 3 designer to convert type 1 definitions to type 3 definitions. Batteries not included—manual work is required.