A Field Guide to Music Formats

FLAC is the most widely used open lossless codec, and the workhorse of serious music collections. It compresses PCM audio to roughly 50–70% of its original size while guaranteeing that decoding reproduces the source bit-for-bit — no information is discarded, unlike MP3 or AAC. That makes it ideal for archival rips and for serving high-resolution masters.

Development began with Josh Coalson in 2000; version 1.0 was released on 20 July 2001. The project came under the umbrella of the Xiph.Org Foundation in January 2003, the same non-profit behind Vorbis and Opus. FLAC uses linear prediction plus Rice coding of the residual, and supports streaming, seeking, embedded cue sheets and rich metadata via Vorbis comments.

FLAC's standing as a preservation format is formal: it is a recommended format for digital-audio archiving, and in late 2024 the format was published by the IETF as RFC 9639, giving the long-stable bitstream an official specification.

ALAC is Apple's answer to FLAC: a lossless codec that perfectly reconstructs the original PCM while typically halving the file size. It was introduced in 2004 for use with iTunes and iPod, and ALAC data is normally carried in an MP4 container with the .m4a extension — the same wrapper used for AAC, which is why both share that extension.

For its first seven years ALAC was proprietary. In October 2011 Apple released the encoder and decoder as open source under the Apache License 2.0, which let third-party tools and players adopt it freely. The codec supports up to 8 channels at 16, 20, 24 and 32-bit depths and high sample rates, and it is the format behind Apple Music's lossless tier.

Because the algorithm and container differ from FLAC, ALAC and FLAC are not interchangeable byte-for-byte, but both are lossless — converting between them does not degrade audio.

WAV and AIFF are the two classic uncompressed containers — the raw, un-encoded PCM that lossless codecs compress and that lossy codecs throw data away from. They store sound as linear pulse-code modulation samples, exactly the representation on a Compact Disc (44.1 kHz, 16-bit, stereo), with no compression and therefore no quality loss and no decoding cost.

AIFF (Audio Interchange File Format) was developed by Apple in 1988, derived from Electronic Arts' Interchange File Format (IFF) chunk structure; it became the native uncompressed format on the Macintosh. WAV (Waveform Audio File Format) followed in 1991 from Microsoft and IBM, built on the closely related RIFF (Resource Interchange File Format) container, and became the standard uncompressed format on Windows.

Both are chunk-based and broadly equivalent in audio terms — the difference is mainly byte order and ecosystem. A practical limit is that the classic 32-bit size field caps a file at 4 GiB (about 6.8 hours of CD-quality stereo); professional variants like RF64 and BWF extend this for broadcast use.

WavPack is a free, open-source lossless codec notable for a feature few others offer: a hybrid mode. In hybrid mode it splits the encode into a small, relatively low-bitrate lossy file plus a separate "correction" file; the lossy part plays on its own, and adding the correction file restores perfect lossless quality. This lets one master serve both a compact everyday copy and a bit-perfect archive.

Development was started by David Bryant in mid-1998, with version 1.0 released on 15 August 1998. The format is maintained under a BSD license and continues to see active releases. Alongside hybrid mode it supports standard pure-lossless compression, high bit depths and sample rates, and multichannel audio.

WavPack is less ubiquitous than FLAC but is valued by archivists and tinkerers precisely for that hybrid split, which no mainstream lossless format replicates.

MP3 is the format that made digital music portable. It is a lossy perceptual codec: it uses a psychoacoustic model to discard sound the human ear is unlikely to notice (masked frequencies, quiet detail behind loud passages), achieving roughly 10:1 compression over CD audio. The trade-off is that the discarded information is gone for good — unlike FLAC, an MP3 cannot be decoded back to the original waveform.

It was developed largely at the Fraunhofer Society in Germany under Karlheinz Brandenburg, with the underlying standard approved in 1991, finalized in 1992 and published in 1993 as ISO/IEC 11172-3 — MPEG-1 Audio Layer III. Layer III was the most complex and most efficient of the three MPEG-1 audio layers.

Through the late 1990s and 2000s MP3 became synonymous with digital music, file sharing and the first portable players. Its core patents have since expired, and while newer codecs are more efficient, MP3 remains universally compatible.

AAC was designed as the successor to MP3, delivering better sound quality at the same bitrate (and acceptable quality at lower bitrates). Like MP3 it is lossy and perceptual, but it uses larger transform blocks, more sample frequencies, and more sophisticated coding tools, which is why a 256 kbps AAC generally sounds better than a 256 kbps MP3.

It was standardized by the Moving Picture Experts Group, first as part of MPEG-2 in 1997 and later extended under MPEG-4. Development was a collaboration among Fraunhofer IIS, Dolby, AT&T and Sony — Fraunhofer again, the same lab behind MP3. AAC reached mass adoption as the default format of Apple's iTunes Store and iPod, and it is the audio in most MP4 video and streaming today.

AAC streams are usually stored in MP4/.m4a or .aac files. SoniqBoom treats .m4a containers as either AAC (lossy) or ALAC (lossless) depending on the codec inside.

Vorbis is a free, open, patent-unencumbered lossy codec created as an alternative to the then patent-encumbered MP3 and AAC. It typically matches or beats MP3 at a given bitrate and was an early favorite for open-source software, games and streaming where licensing fees were unwelcome. The codec is properly called Vorbis; "Ogg Vorbis" refers to Vorbis audio inside the Ogg container, which is its usual home (.ogg / .oga).

It is a project of the Xiph.Org Foundation — the same group behind FLAC and later Opus. The first packaged release came in 2000 and version 1.0 stabilized in July 2002. Vorbis introduced the Vorbis comment metadata scheme that FLAC also adopted.

Xiph now recommends Opus for new lossy encoding, but Vorbis remains widely supported and is still common in game audio and older streaming catalogs.

Opus is the current state of the art in open lossy audio: it handles everything from low-bitrate speech to full-quality music in a single codec, with very low latency, and is royalty-free. It works well across an enormous bitrate range and is the default audio codec for WebRTC, Discord-style voice, and much of YouTube and modern streaming.

It was developed by the Xiph.Org Foundation with contributors from Skype and Mozilla, combining two technologies — SILK (from Skype, strong on speech) and CELT (low-latency music). Key developers included Jean-Marc Valin, Koen Vos and Timothy B. Terriberry. The final specification was published as RFC 6716 on 10 September 2012, making Opus an IETF standard.

For music delivery Opus generally outperforms Vorbis and AAC at equal bitrates, and like its Xiph siblings it carries no licensing burden.

DSD (Direct Stream Digital) is a fundamentally different way of representing sound from everything above. Instead of multi-bit PCM samples taken tens of thousands of times per second, DSD uses a stream of single-bit values via pulse-density modulation — the proportion of 1s to 0s, sampled at 2.8224 MHz (64× the CD rate, "DSD64"), encodes the waveform. Higher rates like DSD128 and DSD256 push that further.

Direct Stream Digital is a trademark of Sony and Philips, created for the Super Audio CD (SACD), introduced in 1999 as a high-resolution successor to the CD. SoniqBoom handles DSD in two file forms: .dsf (DSD Stream File, Sony's format, which carries tags) and .dff (DSDIFF / DSD Interchange File Format, Philips', common in raw SACD rips).

Because most playback chains and cast targets are PCM, the server transcodes DSD to high-resolution PCM for delivery while preserving the source files untouched.

A SID file is not recorded audio — it is 6502 machine code plus a small player routine that, when executed, drives the Commodore 64's sound chip. To play one, SoniqBoom emulates the chip and the CPU in real time, which is why the same file sounds slightly different on different emulation cores.

The chip is the MOS Technology 6581, universally known as the SID (Sound Interface Device), designed by Bob Yannes and shipped in the Commodore 64 in 1982. It was a genuine single-chip synthesizer: three oscillators with multiple waveforms, ADSR envelopes, ring modulation and a famously characterful analog multimode filter. A later revision, the 8580 (around 1986), used a different process and had subtly different, cleaner filter behavior — so collectors distinguish 6581 vs 8580 recordings.

The definitive archive is the High Voltage SID Collection (HVSC) at hvsc.c64.org, holding tens of thousands of tunes. Two companion datasets matter for playback: Songlengths (an MD5-keyed list giving the correct duration of each subtune, since the code itself never "ends") and STIL, the SID Tune Information List, which adds composer credits, cover-song notes and trivia.

A MIDI file contains no sound at all — it is a score: a timed list of note-on / note-off events, instrument selections and controllers. The actual audio depends entirely on the synthesizer rendering it. MIDI (Musical Instrument Digital Interface) was standardized in 1983 by a panel of synth makers, famously demoed by Dave Smith at NAMM, to let instruments from different companies talk to each other.

Because a raw MIDI file could sound wildly different per device, General MIDI (GM) was published in 1991 by the MIDI Manufacturers Association and the Japan MIDI Standards Committee. It fixes a standard 128-instrument patch map (program 1 is always Acoustic Grand Piano, etc.) and minimum polyphony, so a file plays recognizably everywhere.

To render GM consistently, SoniqBoom uses a SoundFont. The .sf2 format stores banks of recorded instrument samples plus instructions for pitching and looping them — sample-based synthesis driven by the MIDI score. SoundFont was developed by E-mu Systems and Creative Labs, first used on the Sound Blaster AWE32 in 1994, and is now an open standard, so any GM-compliant .sf2 can be swapped in to change the sound.

NSF (NES Sound Format) packages the music engine and data extracted from a Nintendo Entertainment System game, along with the 6502 code that plays it. Like SID, it is executable music: the player emulates the NES CPU and its audio hardware to reproduce the tune, and exposes the game's separate tracks as selectable subtunes.

The sound comes from the audio unit inside the Ricoh 2A03 (the 2A07 in PAL regions), the NES's CPU, which integrates the APU (Audio Processing Unit) alongside a 6502-derived processor. The APU provides five channels: two pulse/square waves, a triangle wave, a noise channel and a DPCM sample channel — the unmistakable palette of NES music. Some games added expansion sound chips on the cartridge (Konami VRC6, Nintendo's FDS, Namco 163, Sunsoft 5B), which NSF can also represent.

SoniqBoom renders NSF through its game-music engine, which emulates the 2A03 APU and the common expansion chips.

An SPC file is a snapshot of the Super Nintendo's entire sound subsystem at the moment a tune is playing: the 64 KB of audio RAM, plus the registers of the sound CPU and DSP. To play it, SoniqBoom loads that memory image into an emulation of the SNES audio hardware and lets it run — so an SPC is closer to a frozen machine state than to a sequence of notes.

The SNES audio system was designed by Ken Kutaragi and built by Sony, running independently of the main console. It pairs an 8-bit SPC700 CPU with a 16-bit S-DSP that mixes eight channels of compressed PCM samples (BRR format) with hardware ADSR envelopes and, distinctively, a built-in echo/reverb unit — which is why SNES soundtracks have that lush, ambient character.

Because the SNES used real recorded samples rather than pure tone generators, SPC music can sound surprisingly rich for 1990 hardware.

GBS (Game Boy Sound) captures the music driver and data from a Game Boy title, played back by emulating the handheld's audio hardware. As with NSF and SID, the file holds code and data rather than a recording, and exposes a game's tracks as subtunes.

The original Game Boy launched from Nintendo in 1989 (developed under Gunpei Yokoi's Nintendo R&D1). Its sound hardware, integrated into the system's Sharp LR35902 core, offers four channels: two pulse/square-wave channels (one with a frequency sweep), a 4-bit programmable wavetable channel, and a noise channel. That wavetable voice — a user-definable 32-step waveform — gives Game Boy music its signature buzzy, hollow timbres beloved by chiptune artists.

The Game Boy's sound chip remains a living instrument: trackers like LSDj run on real hardware, and GBS preserves the original game scores in the same idiom.

VGM takes a different approach from the executable formats above: rather than storing CPU code, it is a log of the exact register writes sent to the sound chips, timestamped sample-by-sample. Playback re-issues those writes to emulated chips, so a VGM is essentially a recording of chip commands — accurate to the original hardware regardless of the game's CPU.

This makes VGM a near-universal container for chip music. A single file can target many systems and their sound chips: the Sega Master System, Game Gear and Mega Drive/Genesis (Texas Instruments SN76489 PSG and Yamaha YM2612 FM), the MSX, Neo Geo, and a long list of arcade boards. Files use the .vgm extension, or .vgz when gzip-compressed.

Because it logs every write at hardware resolution, VGM is prized for fidelity: it reproduces the precise timing quirks of FM and PSG chips that higher-level formats might smooth over.

AY files hold music for the General Instrument AY-3-8910 programmable sound generator (and its compatible Yamaha YM2149 variant), the chip that gave 8-bit European home computers their voice. They are most associated with the ZX Spectrum 128 and the Amstrad CPC, and the format also accommodates the Z80 code that drives the tunes.

The AY-3-8910 was designed by General Instrument in 1978 as a 3-voice PSG: three square-wave tone channels plus a shared noise generator and a single envelope generator, covering nearly the full range of a piano. It was enormously popular, also appearing in the Intellivision, Vectrex, Atari ST, MSX machines and countless arcade cabinets, which is why its bright, buzzy three-channel sound is instantly recognizable.

SoniqBoom renders AY tunes by emulating the PSG (and the host CPU where needed), reproducing the chip's distinctive envelope-driven timbres.

KSS files store music ripped from MSX computer software, bundled with the Z80 code that plays it. The MSX was a standardized home-computer architecture launched in 1983, hugely popular in Japan, the Netherlands, Brazil, Korea and parts of the Middle East, and home to early work by composers who later shaped console game music.

What makes KSS interesting is the breadth of MSX sound hardware it can drive. The base machines used the AY-3-8910 PSG (here called the PSG). MSX-MUSIC added the Yamaha YM2413 (OPLL) FM chip, MSX-AUDIO used the Y8950, and Konami cartridges shipped their own SCC wavetable chip — famous from games like Gradius and Nemesis. A KSS file can call on these in combination.

SoniqBoom's game-music engine emulates the MSX CPU and these sound chips so KSS tunes play with the right mix of PSG, FM and SCC voices.

SAP (Slight Atari Player) is the standard format for music from the Atari 8-bit computers (400/800 and the XL/XE line). It carries the 6502 player code and song data, played back by emulating the Atari CPU and its sound chip.

That chip is the POKEY (Pot Keyboard Integrated Circuit), designed by Doug Neubauer at Atari and first released with the Atari 400 and 800 in 1979. POKEY is a multi-purpose part — it scanned the keyboard, read paddle controllers and handled serial I/O — but for music it provides four audio channels with a flexible set of polynomial-counter "distortion" modes that yield everything from clean tones to gritty noise. Channels can be paired for higher-resolution pitch, and the same chip later powered Atari arcade games.

Some SAP tunes assume a machine with two POKEYs (stereo); SoniqBoom's emulation supports the dual-POKEY case where the file calls for it.

This family is the sound of early-1990s PC gaming: FM synthesis on the Yamaha YM3812, better known as the OPL2. Rather than playing samples, FM synthesis generates timbres mathematically by having operators modulate each other's frequency — a technique Yamaha popularized with the DX7 synthesizer and miniaturized into cheap sound-card chips.

The OPL2 (YM3812) was released by Yamaha in 1985 as part of the OPL series (the later OPL3/YMF262 of 1990 doubled the channels and added 4-operator voices). It offers 9 channels of 2-operator FM with four waveform shapes. The chip became the de-facto PC audio standard inside the Ad Lib card (from Ad Lib, Inc., founded by Martin Prevel, 1987) and then the Creative Sound Blaster, which kept OPL compatibility.

SoniqBoom plays this music two ways. Raw register logs and tunes are rendered through an OPL emulator. It also supports IMF (id Music Format), the timestamped OPL2 command stream id Software used for the music in Commander Keen, Wolfenstein 3D and the early Doom-era tools — a format that, like VGM, simply logs the writes to feed the FM chip.