Skip to content

Devlog #020 - The Sound of 1992

An Unidentified File

Every Darklands session opens the same way: a .PAN image sequence, a slow pan across a medieval battlefield. It has always been clear that something was supposed to go with it. OPENDARK.DGT sits next to OPENDARK.PAN in the game directory, and the name is not subtle. DGT. Digital.

Until now, nobody had formally documented what was in it.

The Format

.DGT turns out to be the simplest possible digital audio format: a raw byte stream, no header, no metadata, no compression. Every byte is one audio sample. Unsigned 8-bit PCM, mono. Silence sits at 128; the signal swings above and below from there.

Offset    Size    Description
0x0000    N       Raw PCM samples: one byte each, unsigned, silence ≈ 128

That is the entire format specification. There is nothing else in the file.

Finding the Sample Rate

Because there is no header, the sample rate has to come from somewhere else: either the engine hardcodes it, or it reads it from a config structure not in the file itself. The engine side has not been traced yet, so the rate was determined experimentally: convert the raw bytes to WAV at several candidate rates and compare.

At 8000 Hz, OPENDARK.DGT plays back as recognisable, natural-sounding audio. At 11025 Hz, the same audio is noticeably faster and higher-pitched, plausible for a slightly different hardware configuration, but not the best match. 8000 Hz is consistent with one of the standard Sound Blaster DSP playback rates for the era.

The confirmed working conversion:

ffmpeg -f u8 -ar 8000 -ac 1 -i OPENDARK.DGT OPENDARK.wav

Or in Python:

import numpy as np, wave

data = np.fromfile("OPENDARK.DGT", dtype=np.uint8)
with wave.open("OPENDARK.wav", "wb") as f:
    f.setnchannels(1)
    f.setsampwidth(1)
    f.setframerate(8000)
    f.writeframes(data.tobytes())

Where It Fits in the Engine

The presentation system pairs .DGT with .PAN files. The engine loads both, then plays them in sync through a presentation loop, with audio streaming alongside image sequence rendering. The likely hardware path is Sound Blaster DAC output via DMA-based buffer feeding, driven by a timer, which was the standard playback mechanism for raw PCM in 1992 DOS games.

For the rewrite, the format itself requires no parsing at all:

byte[] raw = File.ReadAllBytes("OPENDARK.DGT");
float[] samples = new float[raw.Length];
for (int i = 0; i < raw.Length; i++)
    samples[i] = (raw[i] - 128) / 128f;

The complexity is entirely in the engine layer: sample rate selection, hardware abstraction, buffer timing. The format is trivial. This is actually good news: it means the audio data will survive a rewrite intact, and the only work is reimplementing the playback driver.

What Remains Open

Three questions about the engine side are still open:

  1. What exact sample rate does the engine use? 8000 Hz is the best experimental match, but confirming it requires tracing the sound subsystem in DARKLAND.EXE.
  2. Is the rate fixed globally or per file? If fixed, one constant needs to be found. If per-file, there must be a lookup table or config structure somewhere.
  3. How is playback implemented? Blocking read? DMA streaming? Interrupt-driven? The audio driver layer is a forthcoming RE target.

The format documentation is now live in the File Formats section of this site.