MPEG-1 Audio Layer 3, more commonly referred to as MP3, is a popular digital audio encoding and lossy compression format, designed to greatly reduce the amount of data required to represent audio, yet still sound like a faithful reproduction of the original uncompressed audio to most listeners. It was invented and standardized in 1991 by a team of engineers directed by the Fraunhofer Society in Erlangen, Germany.
Overview:
MP3 is a compression format. It provides a representation of pulse-code modulation-encoded (PCM) audio data in a much smaller size by discarding portions that are considered less important to human hearing (similar to JPEG, a lossy compression for images).
A number of techniques are employed in MP3 to determine which portions of the audio can be discarded, including psychoacoustics. MP3 audio can be compressed with different bit rates, providing a range of tradeoffs between data size and sound quality.
The MP3 format uses a hybrid transformation to transform a time domain signal into a frequency domain signal:
32-band polyphase quadrature filter
36 or 12 tap MDCT; size can be selected independent for sub-band 0...1 and 2...31
Aliasing reduction postprocessing
In terms of the MPEG specifications, AAC (Advanced audio coding) from MPEG-4 is to be the successor of the MP3 format, although there has been a significant movement to create and popularize other audio formats. Nevertheless, any succession is not likely to happen for a significant amount of time due to MP3's overwhelming popularity. MP3 enjoys extremely wide popularity and support, not just by end-users and software but by hardware such as portable media players (referred to as MP3 players) DVD and CD players.
History
[edit]
Development
MPEG-1 Audio Layer 2 encoding began as the Digital Audio Broadcast (DAB) project managed by Egon Meier-Engelen of the Deutsche Forschungs- und Versuchsanstalt für Luft- und Raumfahrt (later on called Deutsches Zentrum für Luft- und Raumfahrt, German Aerospace Center) in Germany. This project was financed by the European Union as a part of the EUREKA research program where it was commonly known as EU-147. EU-147 ran from 1987 to 1994.
In 1991, there were two proposals available: Musicam (known as Layer 2), and ASPEC (Adaptive Spectral Perceptual Entropy Coding). The Musicam technique, as proposed by Philips (The Netherlands), CCETT (France) and Institut für Rundfunktechnik (Germany) was chosen due to its simplicity and error robustness, as well as its low computational power associated to the encoding of high quality compressed audio. The Musicam format, based on sub-band encoding, was a key to settle the basis of the MPEG Audio compression format (sampling rates, structure of frames, headers, number of samples per frame). Its technology and ideas were fully incorporated into the definition of ISO MPEG Audio Layer I and Layer II and further on of the Layer III (MP3) format. Under the chairmanship of Professor Mussmann (University of Hannover) the editing of the standard was made under the responsibilities of Leon van de Kerkhof (Layer I) and Gerhard Stoll (Layer II).
A working group consisting of Leon Van de Kerkhof (The Netherland), Gerhard Stoll (Germany), Yves-François Dehery (France), Karlheinz Brandenburg (Germany) took ideas from Musicam and ASPEC, added some of their own ideas and created MP3, which was designed to achieve the same quality at 128 kbit/s as MP2 at 192 kbit/s.
All algorithms were finalized in 1992 as part of MPEG-1, the first standard suite by MPEG, which resulted in the international standard ISO/IEC 11172-3, published in 1993. Further work on MPEG audio was finalized in 1994 as part of the second suite of MPEG standards, MPEG-2, more formally known as international standard ISO/IEC 13818-3, originally published in 1995.
Compression efficiency of encoders is typically defined by the bit rate because compression rate depends on the bit depth and sampling rate of the input signal. Nevertheless, there are often published compression rates that use the CD parameters as references (44.1 kHz, 2 channels at 16 bits per channel or 2x16 bit). Sometimes the Digital Audio Tape (DAT) SP parameters are used (48 kHz, 2x16 bit). Compression ratios with this reference are higher, which demonstrates the problem of the term compression ratio for lossy encoders.
Karlheinz Brandenburg used a CD recording of Suzanne Vega's song Tom's Diner to assess the MP3 compression algorithm. This song was chosen because of its softness and simplicity, making it easier to hear imperfections in the compression format during playbacks. Some have taken to jokingly refer to Suzanne Vega as "The mother of MP3". Some more serious and critical audio excerpts (glockenspiel, triangle, accordion, ...) were taken from the EBU V3/SQAM reference compact disc and have been used by professional sound engineers to assess the subjective quality of the MPEG Audio formats.
MP3 goes public
A reference simulation software written in C language known as ISO 11172-5 was developed by the members of the ISO MPEG Audio committee in order to produce bit compliant MPEG Audio files (Layer 1, Layer 2, Layer 3). Working in non real time on a number of operating systems it was able to demonstrate the first real time hardware decoding (DSP based) of compressed audio. Some other real time implementation of MPEG Audio encoders were available for the purpose of digital broadcasting (radio DAB, television DVB) towards consumer receivers and set top boxes.
Later on, on July 7, 1994 the Fraunhofer Society released the first software MP3 encoder called l3enc. The filename extension .mp3 was chosen by the Fraunhofer team on July 14, 1995 (previously, the files had been named .bit). With the first real-time software MP3 player Winplay3 (released September 9th, 1995) many people were able to encode and playback MP3 files on their PCs. Because of the relatively small hard drives back in that time (c.500 MB) the technology was essential to store music for listening pleasure on a computer.
MP2, MP3 and the Internet
In October 1993, MP2 (MPEG-1 Audio Layer 2) files appeared on the Internet and were often played back using the Xing MPEG Audio Player, and later in a program for Unix by Tobias Bading called MAPlay, which was initially released on February 22nd, 1994 (MAPlay was also ported to Microsoft Windows).
Initially the only encoder available for MP2 production was the Xing Encoder, accompanied by the program CDDA2WAV, a CD ripper that transforms CD audio tracks to Waveform Audio Files.
The Internet Underground Music Archive (IUMA) is generally recognized as the start of the on-line music revolution. IUMA was the Internet's first high-fidelity music web site, hosting thousands of authorized MP2 recordings before MP3 or the web was popularized.
In the first half of 1995 through the late 1990s, MP3 files began flourishing on the Internet. MP3 popularity was mostly due to, and interchangeable with, the successes of companies and software packages like Nullsoft's Winamp (released in 1997), mpg123, and Napster (released in 1999). Those programs made it very easy for the average user to playback, create, share, and collect MP3s.
Controversies regarding peer-to-peer file sharing of MP3 files have spread widely in recent years — largely because high compression enables sharing of files that would otherwise be too large and cumbersome to share. Due to the vastly increased spread of MP3s through the Internet some major record labels reacted by filing a lawsuit against Napster to protect their copyrights (see also intellectual property).
Commercial online music distribution services (like the iTunes Music Store) usually prefer other/proprietary music file formats that support Digital Rights Management (DRM) to control and restrict the use of digital music. The use of formats that support DRM is in an attempt to prevent copyright infringement of copyright protected materials, but methods exist to defeat most protection schemes. These methods can be exploited by the computer-savvy to produce unlocked files that can be freely copied. A notable exception is the Microsoft Windows Media Audio 10 format, which has not been cracked. This type of audio file can only be pirated by recording the audio stream as it is played. If a compressed audio file is the desired result, this recorded audio stream has to be recompressed and suffers a loss in quality.
Quality of MP3 audio
Because MP3 is a lossy format, it is able to provide a number of different options for its "bit rate"—that is, the number of bits of encoded data that are used to represent each second of audio. Typically rates chosen are between 128 and 320 kilobit per second. By contrast, uncompressed audio as stored on a compact disc has a bit rate of 1411.2 kbit/s (16 bits/sample à 44100 samples/second à 2 channels).
MP3 files encoded with a lower bit rate will generally play back at a lower quality. With too low a bit rate, "compression artifacts" (i.e., sounds that were not present in the original recording) may appear in the reproduction. A good demonstration of compression artifacts is provided by the sound of applause: it is hard to compress because of its randomness and sharp attacks, therefore the failings of the encoder are more obvious, and are audible as ringing or pre-echo.
As well as the bit rate of the encoded file, the quality of MP3 files depends on the quality of the encoder and the difficulty of the signal being encoded. For average signals with good encoders, some listeners accept the MP3 bit rate of 128 kbit/s and the CD sampling rate of 44.1 kHz as near enough to compact disc quality for them, providing a compression ratio of approximately 11:1. MP3s properly compressed at this ratio can achieve sound quality superior to that of FM radio and cassette tape[citation needed], primarily due to the limited bandwidth, SNR, and other limitations of these analog media. However, listening tests show that with a bit of practice many listeners can reliably distinguish 128 kbit/s MP3s from CD originals[citation needed]; in many cases reaching the point where they consider the MP3 audio to be of unacceptably low quality. Yet other listeners, and the same listeners in other environments (such as in a noisy moving vehicle or at a party) will consider the quality acceptable. Obviously, imperfections in an MP3 encode will be much less apparent on low-end computer speakers than on a good stereo system connected to a computer or -- especially -- using high-quality headphones.
Fraunhofer Gesellschaft (FhG) publish on their official webpage the following compression ratios and data rates for MPEG-1 Layer 1, 2 and 3, intended for comparison:
Layer 1: 384 kbit/s, compression 4:1
Layer 2: 192...256 kbit/s, compression 8:1...6:1
Layer 3: 112...128 kbit/s, compression 12:1...10:1
The differences between the layers are caused by the different psychoacoustic models used by them; the Layer 1 algorithm is typically substantially simpler, therefore a higher bit rate is needed for transparent encoding. However, as different encoders use different models, it is difficult to draw absolute comparisons of this kind.
Many people consider these quoted rates as being heavily skewed in favour of Layer 2 and Layer 3 recordings. They would contend that more realistic rates would be as follows:
Layer 1: excellent at 384 kbit/s
Layer 2: excellent at 256...384 kbit/s, very good at 224...256 kbit/s, good at 192...224 kbit/s
Layer 3: excellent at 224...320 kbit/s, very good at 192...224 kbit/s, good at 128...192 kbit/s
When comparing compression schemes, it is important to use encoders that are of equivalent quality. Tests may be biased against older formats in favour of new ones by using older encoders based on out-of-date technologies, or even buggy encoders for the old format. Due to the fact that their lossy encoding loses information, MP3 algorithms work hard to ensure that the parts lost cannot be detected by human listeners by modeling the general characteristics of human hearing (e.g., due to noise masking). Different encoders may achieve this with varying degrees of success.
A few possible encoders:
LAME first created by Mike Cheng in early 1998. It is (in contrast to others) a fully LGPL'd MP3 encoder, with excellent speed and quality, rivaling even MP3's technological successors.
Fraunhofer Gesellschaft: Some encoders are good, some have bugs.
Many early encoders that are no longer widely used:
ISO dist10 reference code
Xing
BladeEnc
ACM Producer Pro.
Good encoders produce acceptable quality at 128 to 160 Kibit/s and near-transparency at 160 to 192 kbit/s, while low quality encoders may never reach transparency, not even at 320 kbit/s. It is therefore misleading to speak of 128 kbit/s or 192 kbit/s quality, except in the context of a particular encoder or of the best available encoders. A 128 kbit/s MP3 produced by a good encoder might sound better than a 192 kbit/s MP3 file produced by a bad encoder. Moreover, even with the same encoder and resulting file size, a constant bitrate MP3 may sound much worse than variable bitrate MP3.
It is important to note that quality of an audio signal is subjective. Placebo effect is rampant, with many users claiming to require a certain quality level for transparency. Many of these users fail an A/B test and are unable to distinguish files of a lower bitrate. A given bit rate suffices for some listeners but not for others. Individual acoustic perception may vary, so it is not evident that a certain psychoacoustic model can give satisfactory results for everyone. Merely changing the conditions of listening, such as the audio playing system or environment, can expose unwanted distortions caused by lossy compression. The numbers given above are rough guidelines that work for many people, but in the field of lossy audio compression the only true measure of the quality of a compression process is to listen to the results.
If your aim is to archive sound files with no loss of quality (or work on the sound files in a studio for example), then you should use Lossless compression algorithms, currently capable of compressing 16-bit PCM audio to 38% while leaving the audio identical to the original, such as Lossless Audio LA, Apple Lossless, TTA, FLAC, Windows Media Audio 9 Lossless (wma) and Monkey's Audio (among others). Lossless formats are strongly preferred for material that will be edited, mixed, or otherwise processed because the perceptual assumptions made by lossy encoders may not hold true after processing. The losses produced by multiple stages of coding may also compound each other, becoming more evident when the signal is reencoded after processing. Lossless formats produce the best possible result, at the expense of a lower compression ratio.
Some simple editing operations, such as cutting sections of audio, may be performed directly on the encoded MP3 data without necessitating reencoding. For these operations, the concerns mentioned above are not necessarily relevant, as long as appropriate software (such as mp3DirectCut and MP3Gain) is used to prevent extra decoding-encoding steps.
Bit rate
The bit rate is variable for MP3 files. The general rule is that more information is included from the original sound file when a higher bit rate is used, and thus the higher the quality during play back. In the early days of MP3 encoding, a fixed bit rate was used for the entire file.
Bit rates available in MPEG-1 Layer 3 are 32, 40, 48, 56, 64, 80, 96, 112, 128, 160, 192, 224, 256 and 320 kbit/s, and the available sample frequencies are 32, 44.1 and 48 kHz. 44.1 kHz is almost always used (coincides with the sampling rate of compact discs), and 128 kbit/s has become the de facto "good enough" standard, although 192 kbit/s is becoming increasingly popular over peer-to-peer file sharing networks. MPEG-2 and [the non-official] MPEG-2.5 includes some additional bit rates: 8, 16, 24, 32, 40, 48, 56, 64, 80, 96, 112, 128, 144, 160 kbit/s.
Variable bit rates (VBR) are also possible. Audio in MP3 files are divided into frames (which have their own bit rate) so it is possible to change the bit rate dynamically as the file is encoded (although not originally implemented, VBR is in extensive use today). This technique makes it possible to use more bits for parts of the sound with higher dynamics (more sound movement) and fewer bits for parts with lower dynamics, further increasing quality and decreasing storage space. This method compares to a sound activated tape recorder that reduces tape consumption by not recording silence. Some encoders utilize this technique to a great extent.
Non-standard bitrates up to 640 kbit/s can be achieved with the LAME encoder and the --freeformat option, however only few MP3 players can play those files.
Design limitations of MP3
There are several limitations inherent to the MP3 format that cannot be overcome by using a better encoder.
Newer audio compression formats such as Vorbis and AAC no longer have these limitations.
In technical terms, MP3 is limited in the following ways:
Bitrate is limited to a maximum of 320 kbit/s
Time resolution can be too low for highly transient signals
No scale factor band for frequencies above 15.5/15.8 kHz
Joint stereo is done on a frame-to-frame basis
Encoder/decoder overall delay is not defined, which means lack of official provision for gapless playback
Nevertheless, a well-tuned MP3 encoder can perform competitively even with these restrictions.
Encoding of MP3 audio
The MPEG-1 standard does not include a precise specification for an MP3 encoder. The decoding algorithm and file format, as a contrast, are well defined. Implementers of the standard were supposed to devise their own algorithms suitable for removing parts of the information in the raw audio (or rather its MDCT representation in the frequency domain). During encoding 576 time domain samples are taken and are transformed to 576 frequency domain samples. If there is a transient 192 samples are taken instead of 576. This is done to limit the temporal spread of quantization noise accompanying the transient.
This is the domain of psychoacoustics: the study of subjective human perception of sounds.
As a result, there are many different MP3 encoders available, each producing files of differing quality. Comparisons are widely available, so it is easy for a prospective user of an encoder to research the best choice. It must be kept in mind that an encoder that is proficient at encoding at higher bitrates (such as LAME, which is in widespread use for encoding at higher bitrates) is not necessarily as good at other, lower bitrates.
Decoding of MP3 audio
Decoding, on the other hand, is carefully defined in the standard. Most decoders are "bitstream compliant", meaning that the uncompressed output they produce from a given MP3 file will be the same (within a specified degree of rounding tolerance) as the output specified mathematically in the standard document. The MP3 file has a standard format which is a frame consisting of 384, 576, or 1152 samples (depends on MPEG version and layer) and all the frames have associated header information (32 bits) and side information (9, 17, or 32 bytes, depending on MPEG version and stereo/mono). The header and side information help the decoder to decode the associated huffman encoded data correctly.
Therefore, for the most part, comparison of decoders is almost exclusively based on how computationally efficient they are (i.e., how much memory or CPU time they use in the decoding process).
ID3 and other tags
Main articles: ID3 and APEv2 tag
A "tag" is data stored in an MP3 (as well as other formats) that contains metadata such as the title, artist, album, track number or other information about the MP3 file to be added to the file itself. The most widespread standard tag formats are currently the ID3 ID3v1 and ID3v2 tags, and the more recent APEv2 tag.
APEv2 was originally developed for the MPC file format (see the APEv2 specification). APEv2 can coexist with ID3 tags in the same file, but it can also be used by itself.
Volume normalization
As compact discs and other various sources are recorded and mastered at different volumes, it is useful to store volume information about a file in the tag so that at playback time, the volume can be dynamically adjusted.
A few standards for encoding the gain of an MP3 file have been proposed. The idea is to normalize the volume (not the volume peaks) of audio files, so that the volume does not change between consecutive tracks.
The most popular and widely used solution for storing replay gain is known simply as "Replay Gain". Typically, the average volume and clipping information about audio track is stored in the metadata tag.