34

Many songs in the so called "tracker" format quite popular in the early 90s (think .MOD, .S3M, .XM, .IT and the like) seemingly used sound samples from real synthesizers. This sometimes could also be seen in the sample names in the tracker music files featuring brands such as Korg or Roland.

As many of those tracker songs were written by enthusiasts, how did they manage to grab those samples? Did they really plug a synthesizer's audio output into a line input of a sound card and digitize the analog sound? But were audio cards of the early 90s at all capable of capturing analog audio at reasonable quality?

Or did they find a way to somehow dump the keyboard firmware and/or reverse engineer sample banks provided by the keyboard vendor and convert the samples into a format that could be understood by Scream Tracker / Fast Tracker (or whatever tracker music editor they used)?

48

I composed a small number of Amiga tracker modules at around the turn of the 1990s, one of which even enjoyed some popularity. These were made first and foremost for my personal amusement — for exploring music creation and sequencing on a computer — but they also ended up being used in some demoscene productions.

My tools of choice for sequencing them were, at first, Oktalyzer (see here as well), and later ProTracker (see here as well).

For the very first experiments, I used whatever was available to me — instrument samples borrowed from tracker modules made by other people or originating from the famous ST-## instrument sample disks, which were originally meant to be used with Karsten Obarski’s Soundtracker, the forefather of ProTracker.

Then I saw this computer magazine ad for a sound digitizer device for the Amiga. It was a black nondescript plastic box — a bit larger than a pack of cigarettes — which plugged into the printer (parallel) port on the back of the Amiga. It had a single mono RCA input and a trimmer knob for adjusting the input level. There was also a cable going to the joystick port which it used for powering itself. That’s about it.

It was quite similar to this device (more information):

Deluxe Sound Audio Digitzer

The audio editor I mostly used with it was Aegis AudioMaster:

Aegis AudioMaster IV

See this video showing it in use. I never utilized this tool’s built-in sequencing capability, though, but only used it for acquisition and preprocessing the instrument samples for the tracker programs.

I also did some sampling and instrument sample processing in the internal audio editors built into some trackers, but I tended to use AudioMaster the most.

The digitizer was 8-bit, and mono only, but it was a perfect match for the Amiga capabilities since the Amiga sound hardware had four 8-bit sound channels. (Of course, I would later learn that oversampling and better-than-8-bit dynamics during acquisition would yield better results when downsampled to the target format. But at the time, the Amiga hardware was all that I had for working with sound.)

You would not normally have used stereo samples due to the limited disk and memory resources and the limited number of sound channels (polyphony) available in the hardware so the audio digitizer being mono only was not really a hindrance.


So, with the hardware and software in order, I then started digitizing instrument samples from commercial music, usually C cassettes and popular songs. Many songs had e.g. intro sequences or breaks from where you could get great, clean drum or bass samples. I also got some pads and major and minor chords from some songs.

A great deal of effort went into finding suitable loop points for sustained sound without it clicking or popping, especially for the strings and the pads. The audio editors had some tools which made this easier, such as functions for finding zero points in the waveform. But a good result would only come by experimenting and listening to how it sounds.

I then started borrowing keyboards from my friends for the purpose of methodically growing my instrument sample library. I used everything I could get my hands on — from cheap entry-level Casios to actual synthesizers, samplers, and drum machines. A friend of mine had an Ensoniq sampler and an Alesis drum machine which were treasure troves.

In the end, the instrument samples in my sample library came from three sources — other people’s sound tracker modules (some of the instrument samples other people had made were just too good to dismiss), self-digitized from compact cassettes containing commercial music and popular songs, and self-digitized from synthesizers and consumer keyboards. I also digitized my own voice a couple of times.

I can no longer recall the purchase price of the audio digitizer device but I’d estimate it was maybe something like €20 or €30 in today’s money. Certainly not a professional thing at any level but worked fine for the purpose it was made.

There’s this nice video from “debuglive” on YouTube which describes the very process:

https://www.youtube.com/watch?v=i9MXYZh1jcs

| improve this answer | |
  • 1
    Wow, thanks a million for such a detailed write-up! And the guitar sound in your module is the cleanest I believe I've ever heard in 8-bit tracker songs. – DmytroL Aug 5 at 9:38
  • 2
    @DmytroL: Thanks! You can actually listen to the original Ensoniq EPS “Dream Guitar” instrument patch here: youtube.com/watch?v=cM5wFMYQ_cY . I considered inspecting things a bit further and possibly making disk images of those instrument patch floppies and seeing if I can figure out the filesystem and sample format, maybe allowing a direct digital conversion, but never got around to it. – Jukka Aho Aug 5 at 13:56
  • I regret that I have only one upvote to give! Thank you for this detailed answer that was a joy to read. – AJM-Reinstate-Monica Aug 7 at 17:45
27

The four main ways were:

  • you knew someone with the sampling card or sampler (many of samplers had the capability to sample sound in CD quality and save it to floppy disk).
  • you bought some kind of "sample bank"
  • you simply grabbed sounds from other MODs
  • you generated fully artificial sound by math algorithms and computed sample data (e.g. simple sawtooth or sine wave, AM/FM modulated with LFO, filtered, envelopes, etc.) The 90s computers have no power to do it real-time, but it was possible to pre-computed in lower quality and then re-compute in full quality
| improve this answer | |
  • 3
    There was also a lot of shareware at the time, in my country I used to get "ST Magazine" which came with a disk full of free stuff, including music samples (which I reckon were often a bit cheap) – Kaddath Aug 5 at 7:51
  • 2
    I think I remember seeing samples available on BBSes also. – snips-n-snails Aug 5 at 15:13
  • I know a LOT of scene musicians back then (I was a scener too in Magnetic Fields and Digital (DTL)) and I don't know anyone who bought a sample bank as you suggest. Maybe the odd non-scener did from public domain libraries possibly but in general sceners hated PD libraries with a vengeance. We swapped sample disks by post (not many folks had modems then apart from crackers/traders) or we sampled our own from synths or we used sounds from other mods. We also sampled from records from the time (I sampled from the b-side of an SL2 rave track for example for my Pandemonium mod) - Yaz – Rich Campbell Aug 11 at 5:35
17

Did they really plug a synthesizer's audio output into a line input of a sound card and digitize the analog sound? But were audio cards of the early 90s at all capable of capturing analog audio at reasonable quality?

Samplers existed even for low-end machines. I remember using a sampler on my Amiga 500. It was 8-bit, but so are the .MOD samples, and sample rate on the Amiga is limited to 28 kHz per channel too.

On IBM-PCs with a high-end sound card, it was probably possible to get 16-bit/higher sample rate (then convert to 8SVX or whatever).

To create good samples from those keyboards, you had to record the samples with a sampler:

  • using different octaves (different octaves often mean different sounds, plus strecting the sample rate will yield mediocre results). Doesn't apply to drum/percussion samples.
  • for strings or piano, you had to record chords to avoid wasting module channels (often limited to 4, with a stretch at 7 or 8 but with extra CPU power)
  • for strings/leads, you had to make sure the looping point was at the same level as the starting point, else the sample could "click" when looping
  • using different note attacks (if you really wanted that the instrument sounded good)

Very rarely, what was possible was to record from real songs, for instance when the drums/instrument is alone on the track.

I also remember buying an audio CD with a lot of sampled sounds from known keyboards and rythm units. Those could be extracted to .wav and converted, or just sampled (wasteful, but converting to .wav wasn't easy before the CD-ROM era).

Or did they find a way to somehow dump the keyboard firmware and/or reverse engineer sample banks provided by the keyboard vendor and convert the samples into a format that could be understood by Scream Tracker / Fast Tracker (or whatever tracker music editor they used)?

That would be difficult and not practical because a sound is often not a sample but a mix of sample + effects. For instance, if you press a key very strongly, the keyboard could issue another sample or a mix of 2 samples because the attack is different (for instance a piano).

Now MT32 emulators exist (for instance munt) and the MT32-like roms have been dumped, but it's still difficult to convert the samples from this data.

| improve this answer | |
  • 1
    I saw my first sound sampler for the Amiga well before the A500 came out. It was at my local Amiga club in Melbourne Australia. I don't know if the guy designed and made them himself, sold them on behalf of a hardware geek friend who made them, or imported them. He had it hooked up to the first Sony CD Walkman I think I'd seen. Impressed the hell out of me. I had done low res sampling myself on the ZX Speccy but this was the first "perfect" sampling I ever heard when only the Fairlight CMI was known for such quality sampling. – hippietrail Aug 5 at 9:24
  • My first and only sampler on amiga was a "beeline" model, pretty cheap as I remember, coupled to AudioMaster. I may still have it. – Jean-François Fabre Aug 5 at 12:15
  • 1
    From what I read, the joystick ports on the Tandy Color Computer--unlike most analog joystick or paddle ports--were capable of sampling at audio frequencies using software-based successive approximation. They didn't have a sample and hold, so unless an external sample and hold were added, the successive approximation process would add distortion, but the quality would still be better than the speech on the "Let's Count" game I did for the Apple II in the 1980s. – supercat Aug 6 at 15:20
  • 1
    I did a bit of reading and apparently even the original versions of the famous Fairlight CMI used 8-bit samples. – hippietrail Aug 7 at 0:29
3

But were audio cards of the early 90s at all capable of capturing analog audio at reasonable quality?

Kind of yes, but...

There is relatively little "information" in audio above 10kHz, and you can even go lower than that with many sources. You can certainly recognise presence, absence, or distortion in those higher frequencies, but voice and instruments are perfectly recognisable and clear. The sound will be a little "dull", but in the context of 8-bit audio this was never a problem in principle. As with Dr Johnson's dog, the wonder is not that it does it well, but that it does it at all!

With higher frequencies though, they cause aliasing if they go into the sampler. The effect of aliasing is to give a "reflected" frequency from half the sample rate, so if you sampled at 20kHz and you stuck signals at 11kHz, 12kHz and 13kHz into a sampler then you'd record aliased signals at 9kHz, 8kHz and 7kHz. Basically this gives you all sorts of distorted junk you don't want unless you cut hard at half your sample rate.

8-bit sampling also was not a problem in principle. An interesting feature of sampled sound is that its quality depends as much on dither as it does on bit depth. Appropriate dither adds "hiss", but also recreates the signal properly. The UK audio industry magazine SoundOnSound had a great article about digital sampling where they reduced a piano recording to 3-bit sampling, and with the correct dither it was still perfectly audible.

In practise though, one problem for all sampled audio in the 80s was that anti-aliasing was not well understood - or even when understood, the technology did not exist to filter higher frequencies well enough to prevent aliasing. And the other problem was that dither was also not well enough understood, and at 8-bit or even 16-bit depth the resolution is bad enough that you need it. The result was that sampled recordings suffered for the lack of both these features.

This is why early CDs in the 80s got such a bad reputation for their sound. There was nothing in principle wrong with CDs or with CD players, and the myth of CDs and players having any kind of "sound" is provably mathematically wrong. (The truth is that analogue playback media does have a "sound", and what you hear from a CD is accurate playback without the distortion that vinyl or tape will always add.) Where things went wrong in the early days was with the sampling stage, because the electronics of the day didn't have adequate anti-aliasing and the mastering stage generally didn't include dither. This was mostly solved by the 90s, but the myth persisted.

TL;DR: Audio cards were capable of capturing that audio at reasonable quality; but they often should have had an extra analogue filter on the way in, and the recording needed dithering adding. And awareness of those two features back in the 80s was very low. So sampled audio often did not use those audio cards to the full extend of their abilities.

| improve this answer | |
  • Thanks @Graham! It is particularly educating that ADCs (even if 8-bit ones) were commonly available in the 80s so it was economically feasible to use them in consumer hardware. – DmytroL Aug 6 at 14:42
  • I find it somewhat curious that audio capture hardware was obscure enough that programs to play clips of digitized music were popular downloads from services like Quantum Link, even though the circuitry required to capture audio at the quality of such files would have been sufficiently simple and cheap that the cost of the connectors could have exceeded the cost of all the other components combined. Though now I'm almost feeling inspired to try building something. For use with a C64 or C128, the hardware budget might have been reducible to a transistor and some resistors, for... – supercat Aug 7 at 17:05
  • ...4-bit 8kHz quality audio. Program the SID to take the audio input, filter it, mix it with a noise waveform, and output it. Then use the transistor to detect whether the output is above or below a desired threshold. Maybe use two transistors if the response from using one would be too mushy. – supercat Aug 7 at 17:07
  • @supercat The problem is the gap between the Nyquist limit (half sampling rate) and the frequencies you want to keep. The smaller that gap, the harder your filtering needs to be, but harder filtering also screws up the phase which isn't good either. It was digital filtering (in hardware) which really sorted things for ADCs for the tiny gap between 20kHz audio and 22kHz Nyquist. It's also the benefit of 88/96kHz sampling; the sample rate doesn't improve the data, it improves the anti-aliasing. – Graham Aug 7 at 20:05
  • @supercat And by the way, it's also why the SID did such a good electric guitar sound. Most of the higher frequencies on an electric guitar are enharmonic artifacts from distortion, so adding a bunch of potent sources of enharmonic noise is almost a good thing! It's less good for speech and anything like piano which needs to be reproduced more accurately, of course. – Graham Aug 7 at 20:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.