NSF players are supposed to provide RAM at $6000, so that shouldn't cause any problems. But 960 bytes is not very much space for samples, especially if there are several of them. Do you know if that is confirmed by any documents?
(And FlashingYoshi, avoid posting if you don't have anything useful to add.)
There are NO documents. This forum post and mine on the NESdev forums are the only known "documents". That's why I bought the cartridge and I will try to determine more on my own.
BootGod's database page on the Famicom Trainer 3 cart states that it has 960 bytes, but I don't know how he determined that information.
Edit: BootGod replied to the thread to show how he determined the size of the chip with this link.
Again, not much is known about this chip and I don't know if we can even trust this sheet stub.
Edit 2: I found a website that is selling datasheets that actually has this datasheet. I contacted them regarding one of their limited trial memberships in order to obtain the document. I'll release it for free when/if I get the membership.
Got a hold of the Mitsubishi M50805 / M50806 datasheet. While I analyze it I want to make sure other people can look at it and it's now being mirrored.
Well, lidnariq in NESdev forums says that this is basically the equivalent to the Texas Instruments speech synthesizer used in the Speak and Spell machines.
The document says that the chip can be used for voice or melody mode. 5 seconds max of voice or 20 seconds max for melody. The Excitation Circuit is a combination pulse/noise generator. All in all this may be a very interesting chip to play with.
This chip uses a method called PARCOR (or Partial Autocorrelation) for its voice synthesis.
Analysis of the M50805 document chip workflow shows:
A mask ROM (or factory printed) area within the chip (or with the FP version (not being the case of Famicom Trainer 3: Aerobics Studio) being programmable) contains 960 bytes of banked PARCOR vocoded data, which flows to 46 bits (or 5 bytes plus 6 bits) of RAM feeding to the PARCOR decoder ROM. The decoded information then breaks off into two lines: One is the decoded pitch, volume and K-parameter of the PARCOR algorithm in the Parameter Register. The other is the decoded pulse or white noise waveform generated by the Excitation Circuit. These are added into an all-pass lattice filter and they have their phases corrected. Finally they are processed out a DAC (digital to analog converter) and passed down and out the expansion audio pins of the Famicom cart.
There is still more to learn, but I think I'm understanding more about how this works.
A free document describing PARCOR's function is located here:
There are a lot of documents on the Wiley Online Library that describe PARCOR voice synthesis, melody synthesis, and tricks to improve quality. However, this is a pay site. More searching must be done.
Ok using a pure speech coding scheme explains the ROM size, since that can achieve very efficient compression, though it's usually only usable for speech.
Anyway let us know if you find anything interesting!
