This is a little command line tool that allows you to convert files stored in the Korg DW-8000 specific WAV format into much easier to handle MIDI sysex SYX files.
The Korg DW8000 synthesizer is from an era when tape storage for computer data was still the cheapest and therefore most widespread option. Think Commodore C64 or Sinclair Spectrum, which stored programs and data on common audio tapes, in Sinclair's case even on standard audio hardware like a casette recorder that will already have been in the household back in 1982.
As the tape is really just a storage for audio data, nowadays this data is usually kept in WAV format, which is the digitized version of the data. But the Korg DW8000 also supports a full MIDI implmentation, and actually it is much nicer to work with MIDI and its digital serial data than with WAV files.
It took a little reverse engineering exercise, because the tape format of the Korg DW8000 is not the same as the sysex format, but rather - I guess - a simple memory dump of the device. And as RAM was precious, the engineers have squeezed the 51 bytes of sysex data into 31 bytes on tape and probably RAM. They even split up a single data item over 2 storage addresses, which makes the reverse engineering a bit harder than it looked at start.
This tool is implemented in Python, so you will need a working python installation (version 3.6 and newer). Python is available for Windows, Mac OS, and Linux, of course. If you have that, installation of the tool is as easy as installing it from pypi
pip install dw8000_wav2syx-christofmuc
This will install three command line tools, of which you will probably want to use only one. To convert a WAV file into a SYX file that you can e.g. send to your Korg with MIDI-OX, just type
dw8000_wav2syx --store True "Volume 8.wav" Volume8.syx
to create the Volume8.syx file. The --store True part instructs the converter to create sysex write requests between the patches, which turns the sysex file into a proper bank dump. Without these, the file is a list of edit buffer dumps, and if you send the file to the synth, only the last patch sent will be in the edit buffer, all others will have been lost.
WARNING: Sending the file created with --store True to your DW8000 will overwrite all patches, make sure to backup first!
Not all files I found in the Internet could be converted, some have a really bad quality and might have also errors in the audio (some of these old tape drives were really no nicely adjusted) that make my simple program fail.
I am interested in trying to load more complicated error cases, so feel free to contact me and provide me with the WAV files that don't load, maybe there is a chance to scratch the data from the WAV file anyway.
There are two command line switches that you can use to experiment yourself:
- If the file is not reported as "clipped", I use a low pass filter before determining the zero-crossings. Sometimes this low pass filter is doing too much, so you can turn it off by specifying
--lowpass Falseon the command line. - Depending on the noise level, to get a clear zero crossing we assume the signal crosses zero and then raises to some value before it returns again to the other side. The so called hysteresis threshold can be set with the parameter
--threshold 0.05, the default is 0.05 but feel free to play around, sometimes 0.1 or even 0.3 has worked better for some files.
If you want to see what is actually stored on tape, you can run the same conversion process in two steps:
dw8000_wav2bin "Volume 8.wav" Volume8.bin
will create a binary file that contains all the bytes as they are written on the tape. Use a hex editor to see what's in there. The second step just does
dw8000_bin2syx Volume8.bin Volume8.syx
converting the bin file to the sysex representation, effectively completing the transform.
There were many ways to store data on tape back in the 80s, luckily the DW8000 service manual even provided a lot of information on the format.
Scratching the bits from the WAV file is really just measuring the length of the zero crossings of the signal, and classifying the length as either a long (0) or a short (1) rectangle. It won't look very rectangular if you look at the WAV file e.g. with Audacity, but that can be contributed to the low-pass filtering expected on an old audio tape, and should not endanger the conversion.
Once you have the bits, detecting the bytes is simple given they are stored with two start and one stop bit, converting the bitstream into a bytestream.
The much harder problem was the reverse engineering of the memory layout, as the tape data is just a memory dump. I even disassembled the DW8000 firmware while trying to figure it out, but in the end generating test data by creating defined MIDI and looking at the resulting audio file was easier, treating the device as a black box which's behavior can be observed from the outside.
As some substantial work has gone into the development of this, I decided to offer a dual license - AGPL, see the LICENSE file for the details, for everybody interested in how this works and willing to spend some time her- or himself on this, and a commercial MIT license available from me on request. Thus I can help the OpenSource community without blocking possible commercial applications.
All pull requests and issues welcome, I will try to get back to you as soon as I can. Due to the dual licensing please be aware that I will need to request transfer of copyright on accepting a PR.
Christof is a lifelong software developer having worked in various industries, and can't stop his programming hobby anyway.