What Is Speech Coding?

Speech coding is the application of data compression techniques to digitized voice signals to reduce the bit rate required for transmission or storage while preserving sufficient intelligibility and naturalness for the intended communication context. A speech coder, or codec, analyzes frames of the input waveform, typically 10 to 30 milliseconds in duration, extracts a compact parametric description, and transmits those parameters to a decoder that reconstructs a perceptually acceptable output. The field draws on digital signal processing, information theory, perceptual acoustics, and filter theory, and has been a core discipline of telecommunications engineering since the 1960s.

Speech coding is closely related to but distinct from general audio coding. Speech coders exploit specific knowledge of how the human vocal tract produces sound, modeling short-term spectral shape with linear prediction and long-term periodicity with pitch analysis. General audio codecs such as the ISO/IEC MPEG AAC family make no such assumption about the source, instead applying perceptual models of auditory masking to identify components the ear cannot perceive and removing them before quantization. The two approaches converge in modern unified codecs intended for both speech and general audio at wideband quality levels.

Speech Coding Standards and Architectures

The International Telecommunication Union has defined the principal narrowband and wideband speech coding standards deployed in the global telephone network. G.711, operating at 64 kbps with logarithmic PCM quantization, remains the baseline codec for circuit-switched telephony. G.729, a conjugate-structure algebraic code-excited linear prediction (CS-ACELP) coder at 8 kbps, became the dominant standard for voice-over-IP applications because it provides good quality at a fraction of the bandwidth. The 3GPP Adaptive Multi-Rate (AMR) codec family, used in GSM and later mobile generations, adapts its coding rate between 4.75 and 12.2 kbps in response to channel conditions. Details on these and related standards are documented in the ITU-T speech codec registry maintained by VOCAL Technologies.

Audio Coding

Audio coding addresses the broader problem of compressing any acoustic signal, including music, environmental sounds, and mixed content, at quality levels acceptable to the human auditory system. The ISO/IEC Moving Picture Experts Group developed a series of perceptual audio codecs, including MPEG-1 Layer III (MP3) and Advanced Audio Coding (AAC), that apply a psychoacoustic model to identify auditory masking thresholds for each frequency band and allocate quantization bits accordingly. Dolby AC-3 and its successor E-AC-3 apply related methods for surround-sound cinema and broadcast formats. In applications where a single codec must handle both speech and music, such as streaming platforms or WebRTC-based conferencing, the Opus codec developed under IETF standardization combines SILK (a linear-prediction speech coder) and CELT (a transform-based audio coder) in a hybrid architecture covering 6 to 510 kbps. The paper by Jerry D. Gibson on speech coding methods and applications situates audio coding within the broader history of voice compression research.

Vocoders and Voice Activity Detection

Vocoders extend speech coding to very low bit rates, sometimes below 2.4 kbps, by representing the voice signal almost entirely through parametric models of the glottal source and the vocal-tract filter. The mixed-excitation linear prediction (MELP) vocoder standardized for US government secure communications operates at 2.4 kbps and achieves intelligible speech under adverse channel conditions where waveform coders fail. Voice activity detection (VAD) is a complementary technique woven into many codec systems: by classifying each frame as speech or silence, VAD enables discontinuous transmission, suppressing output during pauses and reducing average bit rate and bandwidth consumption. An overview of VAD in the context of the G.729B standard appears in research published through IEEE Xplore on robust VAD for DTX operation.

Applications

Speech coding has applications in a wide range of disciplines, including:

• Mobile telecommunications: AMR and EVS codecs in 2G, 4G, and 5G radio networks
• Voice over IP: G.729 and Opus in enterprise SIP trunks and consumer calling platforms
• Satellite communications: MELP and CELP vocoders for bandwidth-limited government and military links
• Digital broadcasting: speech track coding within MPEG transport streams for television and radio
• Conferencing: wideband and super-wideband codecs in HD voice and video conferencing endpoints