Audio Compression is the process of reducing the amount of digital data required to represent audio while preserving either the exact original signal or an acceptable approximation of it. Digital audio begins as samples of an analogue waveform, with parameters such as sample rate and bit depth determining the uncompressed representation. Compression can be lossless, allowing exact reconstruction of the encoded source, or lossy, allowing some information to be discarded in exchange for a substantially smaller representation.
Lossless techniques exploit statistical and structural redundancy, whereas lossy techniques also exploit characteristics of human hearing to remove or reduce information that is considered less perceptually significant. Audio Compression is essential for music distribution, streaming, podcasts, broadcasting, telephony, games, video production and portable media because uncompressed audio can consume considerable storage and transmission bandwidth.
Uncompressed digital audio has a predictable relationship between sample rate, bit depth, number of channels and duration. A stereo signal sampled at a particular rate with a particular number of bits per sample requires a corresponding number of bits for every sample across both channels. Compression reduces this requirement by finding more efficient representations. Lossless audio coding can predict samples and encode the residual differences using statistical methods, often achieving meaningful reductions without changing the decoded waveform.
The achievable reduction varies with the signal because highly predictable material compresses more effectively than noisy or complex material. Lossy compression can achieve much larger reductions by transforming the signal into representations that are more suitable for perceptual analysis and then quantising or removing components according to their expected audibility.
Human hearing is central to many lossy Audio Compression systems. Psychoacoustic principles describe phenomena such as frequency masking, temporal masking and limitations in sensitivity at different frequencies and sound levels. If a strong sound makes a nearby weaker component difficult or impossible to perceive, an encoder can sometimes represent the weaker component with less precision without producing a large perceived difference. Perceptual encoders therefore analyse spectral content and allocate bits according to models of auditory sensitivity.
This is not equivalent to simply deleting high frequencies, because perceptual importance varies with frequency, signal level, temporal context and neighbouring sounds. The effectiveness of a lossy codec depends on the quality of its psychoacoustic model, transform design, quantisation strategy and bitrate allocation. At sufficiently low bitrates, discarded information becomes increasingly audible through distortion, loss of detail, pre-echo, warbling or other artefacts.
Audio Compression methods commonly transform a time-domain signal into frequency-related components so that the encoder can process different spectral regions separately. Transform coding or filter-bank techniques can represent audio using coefficients associated with different frequency bands or temporal windows. Quantisation reduces the precision of these coefficients, after which entropy coding can represent the resulting symbols efficiently.
Windowing creates an important trade-off between time and frequency resolution because short windows respond more quickly to transient events while longer windows can provide finer frequency discrimination. Modern perceptual codecs may switch or adapt their analysis strategies to deal with transients and sustained signals. The encoded result therefore reflects not simply a smaller version of the waveform but a structured representation designed around the characteristics of both the signal and human auditory perception.
Bitrate is a major practical factor in lossy Audio Compression. A higher bitrate generally provides the encoder with more information capacity, but bitrate alone does not determine quality because codec design, encoder implementation, source material and channel configuration also matter. Constant-bitrate approaches maintain a relatively stable data rate, while variable-bitrate approaches allocate more bits to complex passages and fewer bits to simple passages. Average bitrate can therefore provide a better description of storage requirements for some variable-rate encodings.
Stereo audio may be encoded using techniques that exploit similarities between channels, reducing redundancy while preserving the perceived spatial image. Multichannel systems introduce additional considerations because relationships among channels can be exploited without eliminating important spatial information. The practical choice depends on whether the priority is archival fidelity, distribution size, latency, compatibility or computational efficiency.
Lossless and lossy Audio Compression serve different purposes. Lossless methods are useful when exact signal recovery is required, such as archival masters, production assets or situations where repeated processing may occur. Lossy methods are appropriate when storage or bandwidth constraints dominate and a carefully controlled perceptual loss is acceptable. Re-encoding a lossy file into another lossy representation can introduce additional degradation because the second encoder receives an already altered signal.
For production workflows, a high-quality original or lossless master is therefore generally preferable as the source for multiple delivery versions. Compression also interacts with sample rate and bit depth: changing these properties is a form of signal conversion and should not be confused with compression itself, although they may be performed during the same encoding workflow.
Audio Compression has applications ranging from music streaming and downloadable media to voice communications and embedded systems. Music services require efficient storage and delivery at scale, while voice applications can prioritise extremely low bitrate and low latency. Podcast distribution often uses compressed audio because episode files may be downloaded or streamed by large numbers of listeners. Games compress audio assets to reduce installation size and memory requirements, sometimes using different representations for music, speech and sound effects.
Broadcast systems may use compression to fit multiple audio services within available transmission capacity. Professional production can use lightly compressed or uncompressed formats to preserve flexibility during mixing and mastering, followed by stronger compression for consumer delivery. The appropriate level of compression therefore depends heavily on the operational role of the audio representation.
Audio Compression is sometimes confused with volume reduction, dynamic-range compression or a codec. Reducing file size through data compression is different from dynamic-range compression, which deliberately reduces the difference between loud and quiet parts of an audio signal as an audio-processing effect. A codec is the algorithmic system used to encode and decode compressed audio, whereas compression is the broader process and result.
Lossy Audio Compression provides major savings in storage and bandwidth but can introduce irreversible changes, while lossless compression preserves the original samples but generally achieves smaller reductions. Quality also depends on the entire signal chain: poor recording, clipping, excessive noise or unsuitable sampling can limit the result before compression occurs. Understanding these distinctions is important because a technically small file is not automatically low quality, and a high bitrate does not automatically guarantee faithful reproduction if the source or encoding process is inadequate.
