Converting Scream Tracker 3 S3M Modules to WAV: Architecture and Synthesis

In 1994, Finnish demoscene powerhouse Future Crew released Second Reality—widely hailed as the most influential PC demoscene production of all time. Fueling its unforgettable soundtrack by composers Jussi-Pekka Ilmari ("Purple Motion") and Peter Hajba ("Skaven") was Scream Tracker 3 and its signature music format: .s3m.

While Commodore Amiga tracker modules (.mod) were tied to the Paula sound chip's 4-channel hardware architecture and rigid PAL clock frequencies, Scream Tracker 3 broke free from Amiga limitations. Designed natively for MS-DOS PCs equipped with Sound Blaster, Gravis UltraSound, or AdLib OPL FM synth cards, the S3M format introduced support for up to 32 digital channels, 16-color panning matrices, arbitrary sample playback rates, and sophisticated volume slides.

This forensic guide analyzes the internal byte structure of Scream Tracker 3 files, walks through pattern packing mechanics and 16-byte paragraph pointers, and explains how convrtr's S3M to WAV engine synthesizes chiptune tracker files into pristine 16-bit PCM audio directly in your browser.

The SCRM File Header Specification

Every valid .s3m file starts with a 96-byte primary header:

| Offset (Bytes) | Field Name | Data Type | Forensic Description | | :--- | :--- | :--- | :--- | | 0x00 - 0x1B | Song Title | 28 ASCII chars | Song name, null-padded | | 0x1C | DOS EOF | 1 byte | Always 0x1A (Stops MS-DOS TYPE command from spilling binary) | | 0x1D | File Type | 1 byte | Tracker file type (0x10 = S3M song) | | 0x1E - 0x1F | Reserved | 2 bytes | Padding | | 0x20 - 0x21 | Order Count | 16-bit uint (LE) | Number of pattern order positions | | 0x22 - 0x23 | Instrument Count | 16-bit uint (LE) | Number of sample/instrument parapointers | | 0x24 - 0x25 | Pattern Count | 16-bit uint (LE) | Total unique 64-row pattern chunks | | 0x26 - 0x27 | Flags | 16-bit bitfield | Bit 0: 0 = Standard, Bit 4: Fast volume slides | | 0x28 - 0x29 | Tracker Version | 16-bit uint (LE) | Created tracker version (0x1320 = ST 3.20) | | 0x2A - 0x2B | Sample Format | 16-bit uint (LE) | 1 = Signed samples, 2 = Unsigned samples | | 0x2C - 0x2F | Magic Marker | 4 ASCII chars | Constant "SCRM" (0x53 0x43 0x52 0x4D) | | 0x30 | Global Volume | 1 byte | Master volume (0 to 64) | | 0x31 | Initial Speed | 1 byte | Default tempo ticks per row (typically 6) | | 0x32 | Initial BPM | 1 byte | Default beats per minute (typically 125) | | 0x33 | Master Volume | 1 byte | Master amplification and stereo flag | | 0x34 | Ultraclick Removal | 1 byte | Number of channels to reserve for GUS click reduction | | 0x35 | Default Pan Flag | 1 byte | 252 (0xFC) if custom panning table is appended | | 0x40 - 0x5F | Channel Map | 32 bytes | Configuration for channels 0 to 31 (0xFF = disabled) |

The Parapointer Addressing Mechanism

To optimize file parsing on real-mode x86 processors with 64KB memory segments, Scream Tracker 3 uses parapointers (paragraph pointers) to reference internal data structures.

A parapointer is a 16-bit little-endian integer pointing to a 16-byte aligned memory boundary. To derive the exact absolute byte offset in the file: absoluteByteOffset = parapointer * 16 (or parapointer << 4)

The header contains three consecutive parapointer arrays:

1. Order List: A sequence of bytes designating which pattern index to play next. Order value 255 (0xFF) marks the end of the song, while 254 (0xFE) represents an empty skip marker.
2. Instrument Parapointers: An array of uint16 values pointing to each instrument header.
3. Pattern Parapointers: An array of uint16 values pointing to each compressed pattern block.

Instrument Headers and C4SPD Frequency

Each instrument block begins with a 78-byte sample descriptor:

• Type: 1 for sample instrument, 2 for AdLib FM instrument.
• Length: 32-bit sample length in bytes.
• Loop Start and Loop End: Loop boundary points for sustained notes.
• Volume: Default instrument playback volume (0 to 64).
• Flags: Bit 0 indicates active loop; Bit 4 indicates 16-bit sample data; Bit 5 indicates stereo sample data.
• C4SPD (Middle C Frequency): A 32-bit integer indicating the sample playback rate in Hertz when playing note C-4. While Amiga Paula chips used fixed hardware periods, S3M instruments declare dynamic playback rates (such as 8363 Hz, 11025 Hz, or 22050 Hz), allowing samples to be pitched precisely across any hardware platform.
• Data Parapointer: A 24-bit high-offset parapointer ((highByte << 16) | lowWord) multiplied by 16 to locate the raw PCM sample bytes.

Channel Compression and Pattern Unpacking

S3M patterns contain 64 musical rows. To avoid wasting bytes on empty tracks, rows are compressed using channel-mask packing:

Each row consists of zero or more channel tokens, terminated by a null byte (0x00):

1. Read the first byte b:
   • If b == 0x00: The current row is complete. Increment the row counter and continue to row 64.
2. Extract the channel index:
   • channel = b & 0x1F (Channels 0 to 31).
3. Inspect the bit flags on b:
   • Bit 5 (0x20): Note and instrument data present. Read 2 bytes:
     • Byte 1: Note (0x00 to 0xDF represents octaves 0-9; 0xFE = Key Off, 0xFF = Empty).
     • Byte 2: Instrument (1-based instrument index).
   • Bit 6 (0x40): Volume byte present. Read 1 byte (0 to 64).
   • Bit 7 (0x80): Effect command present. Read 2 bytes:
     • Byte 1: Command letter (e.g., A = Set Speed, B = Order Jump, C = Pattern Break, D = Volume Slide, J = Arpeggio).
     • Byte 2: Info parameter byte.

Synthesizing S3M to 16-Bit Linear PCM

Synthesizing an S3M module into standard audio requires a multi-voice digital mixer running at a fixed output sample rate (e.g., 44,100 Hz):

1. Clock Calculation:
   • tickDurationInSeconds = 2.5 / BPM
   • samplesPerTick = outputSampleRate * tickDurationInSeconds
   • Each row lasts for Speed ticks (default 6).
2. Pitch Calculation:
   • For a given note and C4SPD frequency: frequency = C4SPD * (2 ** ((note - 48) / 12))
   • The sample step increment per output frame is: step = frequency / outputSampleRate
3. Stereo Mixing and Interpolation:
   • Voices accumulate interpolated sample samples based on channel panning (0 = hard left, 7 = center, 15 = hard right).
   • Accumulated floats are clamped and converted to signed 16-bit little-endian words, packaged into a canonical RIFF WAV container.

Pure Client-Side Privacy

Chiptune modules from demoparties, retro games, and personal mod collections can be played and rendered instantly without installing legacy MS-DOS emulators or third-party desktop utilities. Because convrtr runs the entire S3M synthesis engine directly in JavaScript within your browser tab, conversions happen locally at lightning speed with absolute privacy.