What Is Audio Compression?

Audio compression is the process of reducing the storage size or transmission bandwidth of a digital audio signal, either by eliminating redundant information that can be reconstructed exactly on decoding, or by discarding perceptually irrelevant information that a listener cannot detect. The field sits at the intersection of information theory, signal processing, and psychoacoustics. Modern compression allows a piece of music that would occupy roughly 50 megabytes in uncompressed 16-bit stereo PCM form to be stored in 4 to 6 megabytes with no audible degradation, enabling the distribution and storage of massive audio libraries on constrained media and networks.

Audio compression is divided into two categories based on whether the original signal is recoverable after decoding. Lossless compression preserves every sample value exactly; lossy compression achieves far higher ratios by permanently discarding information judged inaudible. The choice between these approaches depends on the application: professional archiving and production demand lossless fidelity, while consumer streaming and broadcasting typically accept lossy compression in exchange for lower bit rates.

Lossy Compression and Perceptual Coding

Lossy audio compression relies on psychoacoustic models that describe what the human ear cannot hear in a given audio context and therefore does not need to store. Two key masking phenomena are exploited: simultaneous masking, in which a loud tone at one frequency suppresses the audibility of quieter sounds at nearby frequencies, and temporal masking, in which a loud event temporarily raises the hearing threshold in the milliseconds before and after it. A transform coder, typically using the modified discrete cosine transform (MDCT), decomposes the audio into frequency-domain coefficients and then allocates bits to each coefficient in inverse proportion to its perceptual significance. Coefficients well below the masking threshold receive few or no bits. The technical analysis in Sound on Sound's explanation of perceptual coding and MP3 compression illustrates how masking thresholds are computed and applied during encoding.

Lossless Compression

Lossless audio compression removes only statistical redundancy in the signal, leaving the sample values mathematically intact. General-purpose compression methods such as DEFLATE (used in ZIP) offer modest gains on audio data because audio samples change rapidly and have low statistical redundancy compared to text. Dedicated audio lossless codecs exploit inter-sample prediction: a linear predictor models the sample value from nearby past samples, and only the small residual between the prediction and the actual value is stored, typically using entropy coding such as Rice coding. The Free Lossless Audio Codec (FLAC) and Apple Lossless (ALAC) are the most common formats in consumer and archival use. FLAC, for example, achieves compression ratios of approximately 2:1 on typical music material while guaranteeing bit-for-bit identical output on decoding. An evaluation of modern audio compression codecs, including both lossy and lossless approaches, appears in a recent survey on the arXiv preprint server examining objective and perceptual quality metrics across multiple formats.

Codec Standards and File Formats

The dominant lossy codec standards are defined by ISO/IEC through the MPEG process. MPEG-1 Audio Layer III (MP3) established perceptual audio coding for consumer use in the 1990s. MPEG-2 AAC (Advanced Audio Coding) improved coding efficiency and added multichannel support. HE-AAC (High Efficiency AAC) further extends compression at low bit rates by adding spectral band replication, which reconstructs high-frequency content from the coded low-frequency signal. Opus, standardized in IETF RFC 6716, is a variable-bit-rate codec designed for internet communications that covers both narrowband speech and full-bandwidth audio in a single format. The Library of Congress format description for MPEG-D Unified Speech and Audio Coding documents how the USAC standard unifies speech and music compression in a single codec, addressing previously separate application domains.

Applications

Audio compression has applications in a wide range of settings, including:

• Music streaming services, delivering audio at 128 to 320 kbps to subscribers on mobile and fixed networks
• Podcast and spoken-word distribution, where speech codecs at 64 to 128 kbps provide adequate quality with low bandwidth
• Digital audio broadcasting, multiplexing multiple channels into a single broadcast spectrum slot
• Video game audio, packing large quantities of sound effects and music into constrained device storage
• Digital audio workstations and archives, storing multi-track recordings in lossless formats to preserve editing flexibility