The EMT-10 is a rather cheap device, but I need these exact sounds for personal reasons. Instead of sampling, I wanted to grab the samples from the chips to have the best possible quality without noise etc.
The device is a PCM rompler with "AWM" technology, only has 12 patches in three variants (default, bright, mellow), is supposed to have 12 bit samples (which is not 100% correct, see below) so I thought it might be similar to the TG-100. But apparently it is not.
The unit has three chips with 8x512k of wave rom data and another 8x128k chip with presets configuration (and probably the firmware code for the CPU is in there, too).
I successfully desoldered and dumped all four chips and I have been able to extract the samples already. The encoding for the samples is 8 bit DPCM which results in about 11-12 bits of sample data. Due to the encoding, some samples have a non-constant DC offset (they drift away) but a simple high pass filter fixes this.
The first wave rom starts with a LUT that specifies the offsets for the samples in 16 byte blocks. Sample length, and loop information (start & length) are part of the wave rom data too and are in front of each sample. There is more information in there but I have failed so far to decode these, I am not sure if I need that data.
- Sample headers (part of the Wave ROMs, in front of each sample)
The firmware/preset rom starts with an address table (uint16s) with 36 entries, which makes sense as the unit has 12 patches with three variants each. But them I'm stuck. There is a table of length 128 for each of these entries, followed by multiple entries (20-30) with two data blocks: One is always 32 uint16s in size, and then another one between 52-68 uint16s. Some of these values seem to be addresses in the ROM, it can clearly be observed when this firmware is loaded as 16 bit in a wave editor. This seem to be addresses of sample start positions, I checked some of them and they seem to match if multiplied by 256.