What Is Audio Coding?

Audio coding is a field of signal processing concerned with the digital representation, compression, and transmission of audio signals in ways that preserve perceptual quality while minimizing the number of bits required. It draws on information theory, psychoacoustics, and transform signal processing to identify and discard audio information that the human auditory system cannot perceive, thereby achieving bit rates far below those of uncompressed pulse-code modulation (PCM) audio. The resulting coded bitstreams are the foundation of music streaming, digital broadcasting, mobile telephony, and video distribution.

The field encompasses both lossy coding, where some information is permanently discarded in exchange for high compression ratios, and lossless coding, where the original sample values are recovered exactly after decoding. Lossy audio coding predominates in distribution applications because its compression ratios of 10:1 to 20:1 are achievable at perceptually transparent quality, while lossless coding, which achieves ratios of roughly 2:1 to 3:1, is used in archival and professional production contexts.

Psychoacoustic Coding Principles

Lossy audio coders exploit two principal properties of human hearing: simultaneous masking and temporal masking. Simultaneous masking occurs when a loud tone at one frequency raises the hearing threshold for nearby frequencies, making quieter components in those frequency regions inaudible and therefore safe to quantize coarsely or discard. Temporal masking means that for a brief period, roughly 5 to 20 milliseconds before and up to 150 milliseconds after a loud transient, quieter sounds are masked and need not be coded at full resolution. A psychoacoustic model, running in parallel with the encoder, computes the masking threshold at each moment in time and allocates bits to spectral components in proportion to their distance from that threshold. The filter banks used to decompose the audio signal into frequency subbands, typically a modified discrete cosine transform (MDCT), are chosen for their frequency resolution and overlap properties that suit this bit-allocation strategy. The tutorial on MPEG audio compression from IEEE Multimedia provides a detailed technical account of how psychoacoustic models are implemented in practice.

Speech Coding

Speech coding addresses the narrower problem of compressing voice signals, typically from 300 Hz to 3,400 Hz for narrowband telephony or up to 7 kHz for wideband speech, to bit rates of 4 to 64 kbps while maintaining intelligibility and speaker recognition. Rather than psychoacoustic masking, speech coders exploit a source model of the vocal tract. Code-excited linear prediction (CELP), standardized in ITU-T G.728 and G.729, models the vocal tract as a linear predictive filter excited by a codebook of excitation vectors; the encoder searches the codebook for the vector that, when filtered, best reproduces the original speech. The Adaptive Multi-Rate (AMR) codec, mandated for GSM and UMTS networks, adjusts its bit rate dynamically from 4.75 to 12.2 kbps based on channel conditions. The MPEG-D Unified Speech and Audio Coding (USAC) standard combines CELP-based speech coding with transform-based audio coding in a single codec at bit rates below 24 kbps.

MPEG Standards and Audio Description

The MPEG family of standards, developed by the Moving Picture Experts Group under ISO/IEC, defines the dominant international audio coding formats. MPEG-1 Audio Layer III (MP3) and MPEG-2 Advanced Audio Coding (AAC) are the most widely deployed lossy formats. MPEG-7, a separate standard, provides a framework for describing audio content with structured metadata, enabling content-based search and classification. Research on high-quality audio coding for multimedia and MPEG standards published on IEEE Xplore traces the evolution of these formats from MPEG-1 through multichannel audio. An accessible technical account of MP3 and AAC compression mechanisms by Karlheinz Brandenburg, who led the MP3 development at the Fraunhofer Institute, describes the algorithmic choices that defined the format.

Applications

Audio coding has applications in a wide range of systems, including:

• Music streaming services, delivering millions of tracks at 128 to 320 kbps over mobile and broadband networks
• Digital audio broadcasting, encoding multiple audio programs within a single multiplex for DAB and HD Radio
• Mobile telephony and VoIP, compressing voice calls to minimize spectrum and bandwidth consumption
• Video game audio engines, storing large libraries of sound effects and music within constrained storage budgets
• Video production and post-production workflows, archiving master recordings in lossless formats such as FLAC or Apple Lossless