MP3

• 1. Introduction

  ›

  MP3, formally known as MPEG-1 Audio Layer III, shrank a CD's worth of music by as much as 95% without most listeners noticing the difference. That single capability rewired the music industry, created a new category of portable devices, and sparked some of the most bitter copyright battles of the internet age. The questions worth asking are how a compression algorithm developed in a German research lab ended up transforming how every person on earth relates to recorded sound, why a song called "Tom's Diner" became its unlikely proving ground, and what the battles over patent ownership reveal about the fragile legal infrastructure underneath a format billions of people used every day.

• 2. The Science of Hearing Less

  ›

  In 1894, the American physicist Alfred M. Mayer reported something counterintuitive: one tone can be made inaudible simply by the presence of another tone at a lower frequency. That observation, combined with decades of subsequent research, became the scientific foundation on which MP3 was eventually built.

  Richard Ehmer described a complete set of auditory curves related to this masking phenomenon in 1959. Between 1967 and 1974, Eberhard Zwicker conducted research into the tuning and masking properties of critical frequency bands, building on earlier foundational work by Harvey Fletcher and his colleagues at Bell Labs.

  The key insight all this research delivered was that the human ear does not hear everything present in an audio signal. Large portions of sonic information are masked, rendered effectively imperceptible by other sounds that occur at the same moment. If an encoder could identify those inaudible components and simply discard them, the file would shrink dramatically with very little perceptible loss.

  Fumitada Itakura at Nagoya University and Shuzo Saito at Nippon Telegraph and Telephone first applied perceptual coding to speech compression in 1966. By 1978, Bishnu S. Atal and Manfred R. Schroeder at Bell Labs had proposed a speech codec that explicitly exploited the masking properties of human hearing. A complementary approach by M. A. Krasner in 1979 reached the same conclusion but was published in an obscure Lincoln Laboratory Technical Report, which meant its influence on the mainstream field was delayed.

  Meanwhile, a separate mathematical tool was maturing in parallel. The discrete cosine transform, proposed by Nasir Ahmed in 1972 and developed with T. Natarajan and K. R. Rao in 1973 with results published in 1974, provided a powerful way to represent audio in the frequency domain. A refinement called the modified discrete cosine transform, proposed by J. P. Princen, A. W. Johnson and A. B. Bradley in 1987, later became a core component of the MP3 algorithm itself. By the time MPEG called for an audio coding standard in December 1988, researchers had spent nearly a century assembling the pieces.

• 3. Brandenburg and the Original Six

  ›

  Karlheinz Brandenburg began working on digital music compression as a doctoral student at Germany's University of Erlangen-Nuremberg in the early 1980s. His focus was on the psychology of perception: specifically, how people actually experience music rather than what the raw waveform contains. He completed his doctoral work in 1989.

  After graduation, Brandenburg worked as a postdoctoral researcher at AT&T-Bell Labs alongside James D. Johnston. That collaboration fed directly into the format that would become MP3. Back in Germany, at the Fraunhofer Institute for Integrated Circuits in Erlangen, he worked with Bernhard Grill and four other researchers who came to be known informally as "The Original Six". In 1990, he became an assistant professor at Erlangen-Nuremberg, continuing his music compression work with scientists at Fraunhofer's Heinrich Hertz Institute before joining the Fraunhofer IIS staff in Erlangen in 1993.

  The direct technical ancestors of MP3 were two codecs: "Optimum Coding in the Frequency Domain", known as OCF, and Perceptual Transform Coding, known as PXFM. These were merged, together with block-switching contributions from Thomson-Brandt, into a codec called ASPEC. ASPEC won the quality competition when proposals were assessed but was rejected by MPEG as too complex to implement.

  A competing proposal, MUSICAM, was simpler and more robust. It had been developed by a consortium including Matsushita, Philips, CCETT in France, and IRT in Germany, and was primarily designed for digital radio broadcasting. MUSICAM's filter bank structure, its frame format, its sample rates, and its 1152-sample framing all eventually carried over into MP3 as structural foundations. A working group that included Brandenburg, Johnston, and several other engineers from both camps then combined the best of both approaches, targeting quality equal to MP2 at 192 kbit/s while achieving that result at only 128 kbit/s.

  To test whether the compression algorithm was actually working, Brandenburg turned to a specific recording. An a cappella version of "Tom's Diner" by Suzanne Vega became his benchmark. Her nearly unaccompanied voice was deceptively difficult: instrumental music had been easier to compress cleanly, but early versions of the format made Vega's voice sound unnatural. Brandenburg listened to it repeatedly through each refinement of the algorithm. He eventually dubbed Vega the "Mother of MP3", and he later met her in person and heard the song performed live.

• 4. From Standard to Street

  ›

  On the 7th of July 1994, the Fraunhofer Society released the first software MP3 encoder, called l3enc. The filename extension .mp3 was chosen by the Fraunhofer team on the 14th of July 1995; before that date, files had been named .bit. With the release of WinPlay3 on the 9th of September 1995, the first real-time software MP3 player for personal computers, encoding and playback of MP3 files became practical for ordinary users.

  Hard drives of that era held only about 500 to 1000 megabytes. Lossy compression was not an aesthetic preference but a practical necessity: fitting multiple albums' worth of music on a home computer as full audio recordings was otherwise impossible.

  The transition from niche format to widespread phenomenon happened partly through an unplanned act of distribution. A hacker known as SoloH found the source code of the "dist10" MPEG reference implementation on servers at the University of Erlangen shortly after it was made available. He wrapped it in a graphical interface that made it easy to rip CDs and convert them to MP3, then spread the software across the internet. That single act effectively launched the era of digital music distribution.

  Nullsoft's audio player Winamp, released in 1997, accelerated the spread further. By 2023, Winamp still reported a community of 80 million active users. In November 1997, the website mp3.com was already offering thousands of tracks created by independent artists at no cost. Windows Media Player added MP3 support in version 5.2. The first portable solid-state digital audio player, the MPMan, was developed by SaeHan Information Systems of Seoul, South Korea, and released in 1998. The Rio PMP300 followed in the same year, despite efforts by the Recording Industry Association of America to suppress the player legally.

  The first large peer-to-peer file sharing network, Napster, launched in 1999. Major record companies argued that free music sharing cut into sales and pursued lawsuits against Napster, which was ultimately shut down and later sold, before returning as a paid streaming service.

• 5. The Patent Wars

  ›

  In September 1998, the Fraunhofer Institute sent a letter to several MP3 software developers stating that a license was required to distribute or sell encoders and decoders. The letter warned that unlicensed products infringed the patent rights of Fraunhofer and Thomson Consumer Electronics, which later became Technicolor. MP3 license revenues administered by Technicolor generated about 100 million euros for the Fraunhofer Society in 2005 alone.

  The consequences for software development were significant. The LAME MP3 encoder project, one of the most widely used open-source encoders, could not legally distribute compiled binaries in countries that enforced software patents. The project's position was that distributing source code amounted to publishing a description of how an encoder could be built, rather than distributing infringing software itself. Unofficial compiled binaries circulated from third-party sources regardless.

  Sisvel S.p.A., a Luxembourg-based company, administered a separate set of MP3-related patents and previously sued Thomson for infringement. Those disputes were resolved in November 2005, with Sisvel granting Thomson a license. Motorola signed with Sisvel in December 2005. In September 2006, German officials seized MP3 players from SanDisk's booth at the IFA show in Berlin after an Italian patents firm won an injunction against SanDisk on Sisvel's behalf. The injunction was reversed by a Berlin judge, then blocked the same day by a second judge from the same court.

  Alcatel-Lucent filed its own litigation against Microsoft in November 2006, asserting patents allegedly inherited from AT&T-Bell Labs. On the 23rd of February 2007, a San Diego jury awarded Alcatel-Lucent 1.52 billion US dollars for infringement of two of those patents. The court subsequently revoked that award, finding that one patent had not been infringed and that the other was co-owned by AT&T and Fraunhofer, both of which had already licensed it to Microsoft. An appeals court upheld that defense ruling in 2008.

  In the United States, the various MP3-related patents expired on dates ranging across 2007 to 2017. The longest-running patent, held by Technicolor, expired on the 16th of April 2017. In the European Union, all relevant patents had expired by 2012 at the latest. Following the US expiration, free and open-source operating systems such as Fedora began shipping MP3 support by default, ending the need for users to seek unofficial third-party packages.

• 6. What Compression Costs and Keeps

  ›

  Ryan Maguire, a sound artist and composer, approached MP3 compression from an unexpected angle: instead of asking what survives the process, he asked what gets discarded. His project "The Ghost in the MP3" isolates the sounds that the algorithm removes. In 2015, he released a track titled "moDernisT", which is an anagram of "Tom's Diner", composed entirely from the audio fragments deleted during MP3 compression of Vega's song. A detailed account of the techniques and the conceptual framework behind the project was published in the 2014 Proceedings of the International Computer Music Conference.

  The trade-off in any lossy compression is between file size and fidelity, and bit rate is the primary control. At 128 kbit/s, a compression ratio of roughly 11:1 against uncompressed CD audio, quality is acceptable for most listeners. At 192 kbit/s the ratio is approximately 7:1. At 320 kbit/s, the maximum allowed by the ISO standard for MPEG-1, the result is closer to the original. A public listening test comparing two early MP3 encoders operating at roughly 128 kbit/s found one scored 3.66 on a five-point scale while the other scored only 2.22, demonstrating that encoder quality matters as much as bit rate.

  Percussive sounds cause particular difficulty. Applause and triangle instruments expose compression artifacts such as ringing and pre-echo at low bit rates. The "Tom's Diner" recording itself presented a subtle technical challenge: the two stereo channels are almost but not completely identical, a property that causes a phenomenon called Binaural Masking Level Depression, which spatially unmasks noise artifacts unless the encoder recognizes and corrects for the situation.

  A test administered by Stanford University Music Professor Jonathan Berger to incoming students over several years found that student preference for MP3-quality music rose each year. Berger noted that the students appeared to prefer the particular sonic character that MP3 compression introduces, which he described as a kind of sizzle. The lossy artifact had, for at least some listeners, become the preferred sound.