Converting MultiTracker (MTM) Modules to WAV: 32-Channel Track Matrix Synthesis

In 1993, Daniel Gold (known as SubSonic in the demoscene group Renaissance) released MultiTracker Editor for MS-DOS. At a time when tracker formats like ProTracker MOD were confined to 4 channels and Scream Tracker 3 S3M stored rigid channel rows, MTM introduced a modular track-matrix architecture supporting up to 32 concurrent channels.

In this technical breakdown, we explore the internal mechanics of the .mtm binary format and how convrtr emulates 32-channel software mixing into uncompressed 16-bit 44.1kHz stereo WAV audio entirely offline.

---

1. The MultiTracker File Architecture

An MTM module begins with a 66-byte master header identifying the song and setting hardware channel parameters:

| Offset | Size | Description | |---|---|---| | 0x00 | 4 bytes | Magic identification: 0x4D 0x54 0x4D 0x10 (MTM\x10) | | 0x04 | 20 bytes | Song title (ASCII null-terminated) | | 0x18 | uint16 LE | Number of saved track blocks | | 0x1A | uint8 | Index of last pattern (total patterns = LastPattern + 1) | | 0x1B | uint8 | Index of last order (total orders = LastOrder + 1) | | 0x1C | uint16 LE | Length of optional embedded ASCII song comments | | 0x1E | uint8 | Number of sample instruments (up to 32) | | 0x21 | uint8 | Active playback channel count (up to 32) | | 0x22 | 32 bytes | Per-channel stereo panning table (0 = left, 15 = right, 7 = center) |

Following the header, sample metadata headers, arrangement order lists, track matrices, pattern indices, comments, and raw PCM sample waveforms are stored sequentially.

---

2. The Modular Track-Matrix Innovation

Traditional tracker formats duplicate note events across channels even when identical musical figures repeat. MultiTracker solved this by decoupling patterns from tracks:

1. Tracks: Independent 64-row monophonic sequences of 3-byte note cells.
2. Patterns: A 32-entry lookup table pointing to track numbers.

A pattern simply references track numbers for channels 0 through 31. If channel 3 and channel 7 play the same bassline or arpeggio, both reference the same track index. Track index 0 represents silence, eliminating empty row overhead.

---

3. Resampling and Stereo Panning

Each note event contains pitch, sample number, and effect data. The note period determines pitch frequency:

Frequency (Hz) = 3546895 / Period

convrtr resamples 8-bit unsigned PCM waveforms to linear 16-bit samples using linear interpolation:

const step = freq / sampleRate;
voice.samplePos += step;

Each voice is positioned in the stereo field using the song's channel panning table, blended across left and right output channels, and clamped to avoid digital distortion.

---

4. 100% Client-Side Private Synthesis

All binary unpacking, track matrix indexing, sample resampling, and RIFF WAVE encoding execute directly within your browser's local sandbox. No data is ever sent to external cloud servers.