A reference survey of media codecs, with the ones WAI registers marked.
Image, video, audio and speech codecs: standards, open-source projects and research papers. An entry that WAI registers shows its capability string, the name a WAI envelope uses to ask a sink for that decode, and where the registration stands.
Filter by modality, era, status, origin or WAI registration. Search by name, body, capability or description. The full list of registered capabilities is in section 5 of the WAI specification; status and dates for the standards and protocols checked so far are on the references page. Contribute to the glossary on the forge.
WAI registration.
Registration names a bitstream and gives it a capability string. Only the mandatory floor (every conforming sink must implement it) and the recommended set (every conforming sink should) are asked of all sinks, and the entries in them say so.
Pulse Code Modulation. Not a codec in the compression sense, but the representation other audio codecs decode back to. The content of WAV, AIFF and most uncompressed audio.
Logarithmic companding PCM as defined in G.711. Mu-law in North America and Japan, A-law elsewhere. Still standard in circuit-switched telephony.
PCM telephony codec, the toll-quality reference for decades. Still standard for circuit-switched telephony and many SIP deployments.
Adaptive Differential PCM. Predictive coding for digital telephony. Standardized for telecom as ITU-T G.726; also used in cordless phones and early game audio.
The fax compression standards: run-length and modified-Huffman coding for 1-bit scanned documents. Carried inside TIFF and PDF.
Flexible Image Transport System. The standard image format of astronomy: multi-dimensional arrays, structured metadata, integer and floating-point pixel types.
Raster format with optional run-length encoding, 24-bit colour and an alpha channel. Simple to parse, and still common in game and visual-effects texture tooling.
X BitMap. The monochrome predecessor of XPM, also stored as valid C source. Used for X11 monochrome cursors and early web bitmaps.
The first digital video coding standard: motion-compensated DPCM, with no DCT. Its development shaped the ITU-T video standards that followed.
Early ADPCM speech coding, succeeded by G.726.
Military low-bitrate speech codec based on linear predictive coding. Highly intelligible, with a synthetic sound.
Run-length-encoded, indexed-colour raster format from the DOS era. Still readable in image libraries.
Icon container. Holds one or more bitmaps (or PNG images, in newer files) at several sizes and colour depths. The usual container for website favicons.
Cursor format. The icon container with added hotspot metadata for the cursor's click point.
Code-Excited Linear Prediction. Not a single codec but the algorithmic family behind most speech codecs deployed from 1990 to 2010.
Tagged Image File Format. More container than codec: it holds several compression schemes (uncompressed, LZW, deflate, JPEG, fax coding). Used in scanning, printing, scientific imaging and archives.
Graphics Interchange Format. LZW-compressed indexed colour with a 256-colour palette and simple animation. Persists for short looping animations.
Module music formats. Not codecs in the usual sense: short sample data plus pattern-sequencing data. The audio formats of demoscene productions.
Portable Bitmap, Graymap, Pixmap and Arbitrary Map. A family of deliberately simple uncompressed formats, used as the intermediate format in Unix image-processing pipelines.
The first practical digital video coding standard, designed for ISDN video telephony. Established the DCT, motion-compensation and entropy-coding template later ITU-T and ISO/IEC video standards build on.
7 kHz wideband ADPCM. An early wideband voice codec, built into many desk VoIP phones.
Regular Pulse Excitation with Long Term Prediction. The algorithm inside GSM-FR.
X PixMap. A text-based image format for icons and cursors, stored as valid C source so images compile directly into programs. Persists in X11 desktop themes.
RGBE format: 8-bit RGB with a shared 8-bit exponent for HDR storage. Still common in image-based lighting and physically based rendering.
Vector Sum Excited Linear Prediction. Used in IS-54 TDMA cellular and GSM-HR.
Device-independent bitmap. An uncompressed raster format with optional run-length encoding. Persists in icon resources and as a simple intermediate format.
ADPCM at several bitrates, the successor to G.721. Used in cordless phones and some satellite telephony.
MPEG-1 Audio Layer I. The simplest of the three MPEG-1 audio layers.
MPEG-1 Audio Layer II. Designed for digital broadcasting; still used in DAB radio and in MPEG-2 transport streams.
Multichannel audio codec of DVD, ATSC broadcast and cinema; standardized by ATSC as A/52 and by ETSI.
The first digital speech codec of the GSM mobile system, using RPE-LTP.
Improved Multi-Band Excitation. The vocoder of the original Project 25 public-safety radio standard.
DCT-based lossy compression for photographs, from the Joint Photographic Experts Group. Supported by essentially every image decoder, three decades after standardization.
wai.image.jpeg In the standard Recommended set. A variant of ADPCM used widely in early multimedia and game audio.
Low-delay CELP, designed for ISDN and videoconferencing audio.
The predictive-coding lossless mode of the original JPEG standard, algorithmically distinct from baseline JPEG. Used in medical imaging (DICOM).
A PDF holding a single image or vector illustration works as an image format. Used for single-page deliverables.
Designed for CD-rate storage. Its companion audio part, Layer III, became MP3.
MPEG-1 Audio Layer III. The format that defined consumer digital audio for two decades, and still supported by essentially every audio decoder.
Digital Picture Exchange. The film-industry standard for scanned motion-picture frames, with 10-bit log encoding. Survives in digital intermediate workflows.
An early lossless audio codec: predictive coding plus Rice coding. Used by live-recording trading communities before FLAC.
Half-Rate GSM codec using VSELP, designed to increase network capacity.
The video codec of DVD and of first-generation digital terrestrial broadcasting, including ATSC. Still in use for broadcast television.
Low-bitrate video for telephony over PSTN lines and early mobile networks. Used for 3G video calls; succeeded by H.264.
Digital Video. An intra-frame DCT codec at a fixed bitrate, used in consumer and prosumer camcorders from the mid-1990s to the mid-2000s.
Transform-domain Weighted Interleave Vector Quantization. An audio coding tool included in MPEG-4 Audio.
Low-bitrate speech codec, used in some video conferencing and VoIP gateways.
CS-ACELP speech codec, long used in business VoIP.
Portable Network Graphics. Lossless, deflate-compressed, with an alpha channel. The Third Edition became a W3C Recommendation in 2025.
wai.image.png In the standard Mandatory floor. Low Complexity Lossless Compression for Images. The research algorithm standardized as JPEG-LS.
Annex A: a reduced-complexity variant of G.729, bitstream-compatible with it.
Mixed Excitation Linear Prediction. Low-bitrate speech coding for military communications.
Enhanced Full Rate GSM codec using ACELP. Deployed in most 2G GSM networks.
Enhanced Variable Rate Codec. The speech codec of CDMA cellular networks.
Advanced Audio Coding, the successor to MP3. A default audio codec of mobile devices, streaming and digital broadcast.
wai.audio.aac In the standard Covers AAC-LC, HE-AAC and HE-AAC v2 in ADTS. Decoder supplied by the sink. Open lossy audio codec derived from MPEG-1 Layer II, aimed at high-bitrate transparent coding.
Block Compression family, BC1 to BC7. Fixed-ratio, block-based GPU texture compression, decoded in GPU hardware.
Open-source lossless audio codec that also supports a hybrid lossless and lossy mode.
Container for block-compressed GPU texture formats (BC1 to BC7). The compressed data uploads to the GPU without a separate decode step.
Low-complexity lossless and near-lossless image coding based on the LOCO-I algorithm. Used where exact preservation matters and compute is limited.
Floating-point HDR raster format (16- and 32-bit). Used in visual effects, ACES colour pipelines and anywhere scene-linear light values must be preserved.
Advanced Simple Profile of MPEG-4 Visual. Implemented by the open-source Xvid project, among others; used on portable devices and digital cameras into the 2010s.
Simple Profile of MPEG-4 Visual. The 3GPP profile for mobile video on early 3G phones; succeeded by H.264 Baseline.
Harmonic Vector Excitation Coding. A very-low-bitrate parametric speech codec in MPEG-4 Audio.
Sub-Band Codec. The mandatory codec of Bluetooth A2DP audio, used unless the devices negotiate a different codec.
Wideband transform codec, used in business video conferencing.
Adaptive Multi-Rate Narrowband. The 3GPP voice codec of 3G calls, still a fallback codec on mobile networks.
Wavelet-based image coding with lossless and lossy modes. Established in medical imaging (DICOM) and digital cinema mastering.
wai.image.jp2 In the standard Cinema mastering class. Released as open source in 2001 and given to the Xiph.Org Foundation as the basis of Theora.
Lossless video codec using predictive coding and Huffman entropy coding. An early lossless intermediate codec for editing, with a family of descendants.
Low-Delay AAC, for real-time two-way audio such as video calls.
Harmonic and Individual Lines and Noise. A parametric audio codec in MPEG-4 Audio.
Lossless audio codec with high compression ratios. Used in archival communities.
Multiple-image Network Graphics. The PNG group's animated format.
Scalable Vector Graphics. An XML-based vector image format, native to web browsers. Used for icons, charts, illustrations and design-tool output.
An open-source implementation of MPEG-4 Part 2.
A wavelet-based video codec project, discontinued before it reached a usable state.
Free Lossless Audio Codec. The common lossless format for music archival and distribution; published by the IETF as RFC 9639 in 2024.
wai.audio.flac In the standard Mandatory floor. Selectable Mode Vocoder, a CDMA voice codec.
General-purpose open audio codec. The audio codec of the original WebM format; succeeded by Opus.
Fast lossless audio codec, still maintained.
Adaptive Multi-Rate Wideband. The standard wideband codec of mobile voice, including VoLTE.
Advanced Video Coding. Hardware decoders for it ship in phones, televisions, set-top boxes and computers, and streaming, conferencing and optical-disc formats carry it.
wai.video.avc In the standard Decoder supplied by the sink. Wavelet-based, visually lossless intermediate codec for editing, standardized by SMPTE as VC-5. An open-source implementation exists.
Range-coded lossless video codec from the FFmpeg project, documented by the IETF as RFC 9043. Used for video archival.
Codec bundled with an open-source screen recorder. Optional delta-frame coding with LZO or deflate.
Video format the ScummVM project uses to repackage cut-scenes from re-engineered games.
High-Efficiency AAC. Adds Spectral Band Replication to AAC. Used in DAB+ and low-bitrate streaming.
wai.audio.aac In the standard Decoder supplied by the sink. Open-source CELP-based speech codec, used in open-source VoIP. Declared obsolete by its developers in favour of Opus.
Internet Speech Audio Codec. A variable-bitrate wideband VoIP codec, released in the open-source WebRTC project and succeeded there by Opus.
Intra-frame codec for broadcast and post-production, standardized by SMPTE as VC-3.
Open-source video codec derived from VP3.
Experimental wavelet-based codec inside FFmpeg, with lossless and lossy modes.
Lossless video codec with arithmetic coding. Used in screen-capture and editing workflows.
HE-AAC plus Parametric Stereo, for very-low-bitrate stereo.
wai.audio.aac In the standard Decoder supplied by the sink. Uncompressed PCM in an Ogg container.
Enhanced AC-3. Higher bitrates and more channels; used for surround sound in streaming.
A lossless audio codec that began as a vendor format and was released as open source in 2011.
Variable-Rate Multimode Wideband, the CDMA counterpart of AMR-WB.
Internet Low Bitrate Codec, designed for packet-loss-resilient internet telephony.
GPU texture compression standardized for OpenGL ES. ETC2 added alpha support.
JPEG 2000 as the picture codec of the Digital Cinema Package, the distribution format for digital cinema.
wai.video.j2k In the standard Cinema mastering class. Super-wideband variant of G.722.1, used in conferencing systems.
A vendor video codec standardized by SMPTE as 421M. A video codec of the HD optical-disc formats.
Audio Lossless Coding. Used in some broadcast archival workflows.
Scalable to Lossless. A lossy AAC base layer that an enhancement layer extends to lossless.
Wideband, scalable extension of G.729, backward compatible with the G.729 base layer.
Successor variants of EVRC, including wideband versions for CDMA voice.
Scalable Video Coding extension to H.264. One layered bitstream; a decoder picks the subset that matches its capabilities. Used in some conferencing and adaptive-streaming systems.
Constrained Energy Lapped Transform. A low-latency music codec, now the transform layer of Opus.
A national audio coding standard, used in domestic broadcast.
A preprocessor that zeroes inaudible low-order bits so that lossless codecs compress audio further.
Animated PNG. A backward-compatible extension: a decoder without animation support shows the first frame. Now part of the PNG Third Edition.
Wavelet-based open-source video codec designed for broadcast use. Its intra-only subset is standardized by SMPTE as VC-2.
A second open-source implementation of Dirac.
Lossless video codec with broad chroma-format support.
Enhanced Low-Delay AAC, with lower latency than AAC-LD. Used in some professional audio-over-IP applications.
Wideband extension of G.711: a backward-compatible enhancement layer.
Embedded scalable codec covering narrowband and wideband speech.
Low-complexity full-band audio codec, used in some conferencing systems.
Began as a vendor format and was standardized as JPEG XR. Supports HDR and wide colour gamut.
Multiview Video Coding. An H.264 extension for stereoscopic 3D video, used on 3D optical discs.
Lossless compression of G.711 streams.
A CELP-derived wideband speech codec, released as open source and adapted as the linear-prediction layer of Opus.
Image format with lossy and lossless modes; the lossy mode is built on VP8 intra coding. Began as a vendor format; documented by the IETF as RFC 9649.
Video codec published as open source and documented as RFC 6386. One of the mandatory-to-implement video codecs for WebRTC browsers (RFC 7742).
Very-low-bitrate speech codec for amateur radio and HF digital voice. Used in FreeDV and several digital amateur-radio modes.
Adaptive Scalable Texture Compression. Variable block-size GPU texture compression covering 2D and 3D textures, HDR and several channel formats.
GPU-decoded video built on block texture compression, for live visuals at high frame rates.
Extended HE-AAC. Combines HE-AAC v2 with USAC's unified speech-and-audio coding. Used for low-bitrate streaming delivery.
Unified Speech and Audio Coding. Combines AAC with speech-coding techniques in one codec. The basis of xHE-AAC.
A hybrid of a linear-prediction speech layer and the CELT transform layer, covering narrowband speech through full-band music. Mandatory to implement in WebRTC.
wai.audio.opus In the standard Recommended set. High Efficiency Video Coding, the successor to H.264. Used in broadcast UHD, UHD optical discs and device capture.
wai.video.hevc In the standard Decoder supplied by the sink. The successor to VP8, with an open-source reference implementation. Hardware decode is common in current devices.
wai.video.vp9 In the standard Decoder supplied by the sink. Open video codec research project. Several of its ideas (lapped transforms, perceptual vector quantization) were carried into AV1.
An open-source JPEG encoder whose output stays decodable by any standard JPEG decoder.
A single-developer image format built on HEVC intra-frame coding.
Audio codec standardized by ETSI as TS 103 190, with object-based audio. Used in ATSC 3.0 broadcast.
wai.audio.ac4 In the standard Immersive audio class. Decoder supplied by the sink. Enhanced Voice Services. The VoLTE successor to AMR-WB, with up to full-band audio.
Backward-compatible HDR extension to JPEG. A baseline JPEG decoder ignores the extension layer; a JPEG XT decoder reconstructs the full dynamic range.
High Efficiency Image File Format, an image container. HEIC is HEIF carrying HEVC-coded images; AVIF is HEIF carrying AV1-coded images.
Free Lossless Image Format, built on MANIAC entropy coding. Development ended when its ideas were carried into JPEG XL.
Limited Error Raster Compression. Bounded-error compression for geospatial raster data. Used in GIS pipelines.
"End-to-end Optimized Image Compression". Established the template of an autoencoder with a learned entropy model, which many later learned image codecs follow.
An autoregressive density model used for lossless image compression, by arithmetic coding over its predicted distributions. A template for many later learned lossless approaches.
"Variational Image Compression with a Scale Hyperprior". Introduced a side-information channel that lets the entropy model adapt to local image statistics.
wai.neural.bmshj2018 In the standard Registers the paper's factorized-prior model. Decode route in the reference sink; model weights supplied by the sink. "Joint Autoregressive and Hierarchical Priors for Learned Image Compression". Added an autoregressive context model on top of a mean-scale hyperprior.
wai.neural.mbt2018_mean In the standard Registers the paper's mean-scale hyperprior model, not its autoregressive context model. Decode route in the reference sink; model weights supplied by the sink. Video codec from the Alliance for Open Media. Hardware decode ships in current devices. Its successor, AV2, was published by AOMedia in 2026.
wai.video.av1 In the standard Recommended set. wai.video.av1.lossless In the standard Recommended set; quantizer 0, lossless in YUV. WaveNet-based low-rate speech coding. An early demonstration that a deep generative model can decode high-quality speech from very-low-bitrate parameters.
Recurrent neural vocoder, smaller and faster than WaveNet. The architecture inside LPCNet and many follow-on works.
Low-complexity, low-latency intra-frame codec for professional video over IP and other uses where latency matters more than compression ratio. Its RTP payload format is RFC 9134.
Image format carrying AV1-coded images. Based on HEIF and conforming to MIAF.
wai.image.avif In the standard Recommended set. Deep Video Compression. An early end-to-end deep video compression framework. Established the residual-coding approach: a predicted frame is generated, and the residual is coded by a learned image codec.
Low-bitrate speech coding with a VQ-VAE and a WaveNet decoder.
Neural speech synthesis through linear prediction. A bridge from classical CELP to neural decoding, practical in real time on a CPU.
GAN-based mel-spectrogram vocoder, often used as the decoder in neural audio pipelines.
Learned image compression with discretized Gaussian mixture likelihoods and attention modules. Widely used as a baseline in later learned-compression papers.
High-Fidelity Generative Image Compression. A GAN-based decoder that produces perceptually convincing reconstructions at very low bitrates, where perceptual quality and pixel fidelity diverge.
Versatile Video Coding, the successor to HEVC.
wai.video.vvc In the standard Decoder supplied by the sink. Low Complexity Enhancement Video Coding. An enhancement layer of residuals that a decoder adds to the upscaled output of a base codec's encode.
wai.video.lcevc In the standard Carried only as a companion to a named base codec. Decoder supplied by the sink. Essential Video Coding, MPEG-5 Part 1. Defined in two profiles: a Baseline profile and a Main profile with further coding tools.
wai.video.evc In the standard Decoder supplied by the sink. wai.video.evc.baseline In the standard Baseline profile only. Decoder supplied by the sink. Multiple-frames Learned Video Compression. Uses several reference frames and a learned in-loop filter.
Low Complexity Communications Codec. The mandatory codec of Bluetooth LE Audio.
GAN-based vocoder producing high-fidelity audio from mel-spectrograms. Widely used in speech synthesis and neural audio research.
Image coding standard with lossy and lossless modes, and lossless transcoding of existing JPEG files. Designed for archival as well as delivery.
wai.image.jxl In the standard Recommended set. Quite OK Image format. A lossless raster format designed to be simple to implement. Used in small-engine game development and embedded contexts.
Feature-domain Video Compression. Moves motion estimation, compensation and residual coding into a learned feature space.
Deep Contextual Video Compression. Moves from residual coding to conditional coding in the feature domain, using learned context from previously decoded frames. The first of a family of follow-on models.
A GAN-based vocoder variant refined for specific audio types.
Open-source neural speech codec for low-bandwidth voice calls. Version 2 is built on a residual-vector-quantized neural codec.
"SoundStream: An End-to-End Neural Audio Codec". Residual vector quantization for speech, music and general audio in one model.
Enhanced Compression Model. JVET's exploration software for coding tools beyond VVC.
Temporal Context Mining variant of DCVC, with multi-scale temporal contexts.
Hybrid Entropy Model variant of DCVC.
Coarse-to-fine motion estimation with hyperprior-guided adaptive coding.
Video Compression Transformer. A transformer-only architecture for learned video compression.
Generative Face Video Coding. A family of methods that send a reference frame plus a stream of facial keypoints or 3D face-model parameters; the receiver animates the face from them.
"High Fidelity Neural Audio Compression". A neural audio codec with residual vector quantization, used as the discrete tokenizer of several open audio-generation models.
wai.neural.encodec32 In the standard The 32 kHz model. Decode route in the reference sink; model weights supplied by the sink. Diverse Contexts variant of DCVC.
Mixed-Motion Video Coding. Chooses a coding strategy per block from its motion characteristics.
Group-residual vector quantization, a refinement of residual vector quantization, for low-bitrate neural audio coding.
"High-Fidelity Audio Compression with Improved RVQGAN". A general-audio neural codec with improved RVQ-GAN training.
wai.neural.dac In the standard The 44.1 kHz model. Decode route in the reference sink; model weights supplied by the sink. A disentangled speech codec whose first codebook is regularized toward phonetic units, separating content from acoustic detail.
A modular neural codec with separately trained encoder, quantizer and vocoder stages, designed to reduce training instability.
A codec designed for downstream language modelling over its tokens.
Perceptual Compression. Diffusion-based image compression at very low bitrates, with realism as the explicit objective.
Feature Modulation variant of DCVC. One model covers a wide quality range, in RGB and YUV.
wai.neural.dcvc_fm In the standard Decode route in the reference sink; model weights supplied by the sink. Deep Hierarchical Video Compression. A hierarchical VAE structure in place of the conditional-coding template.
Unified Intra and Inter-frame Video Compression. One framework handles I-, P- and B-frames.
Audio-Visual Cross-modal Codec. Codes audio and video together at low bitrate, using lip motion to help reconstruct the audio.
Coherent Motion Video Compression. Splits video into semantically coherent clips, each sent as a keyframe plus a motion description that the receiver synthesizes.
A very-low-bitrate transmission scheme: a still image plus a text description, from which the receiver generates the motion.
Multi-Scale Neural Audio Codec. Hierarchical token rates: coarse tokens are sampled less often than fine ones. Used in several open-source speech-synthesis models.
wai.neural.snac In the standard Decode route in the reference sink; model weights supplied by the sink. A frequency-domain neural speech codec toolkit.
A hierarchical codec with separate timbre and content token streams, for voice conversion and speech synthesis.
A streaming neural codec designed for real-time use under varying channel conditions.
A low-frame-rate neural codec designed for audio language models, with a 12.5 Hz token rate.
wai.neural.mimi In the standard Decode route in the reference sink; model weights supplied by the sink. A very-low-bitrate semantic audio codec for general sound, not only speech.
A two-stream codec that separates semantic and acoustic information into different token streams.
Source-Disentangled Codec. Assigns distinct codebooks per audio domain (speech, music, sound effects).
A spectral-domain, delayed-fusion neural codec for low-latency multichannel audio.
A low-frame-rate codec designed as a tokenizer for audio language models.
Learned image coding standard, published in 2025 as ISO/IEC 6048-1 and ITU-T T.840.1. An end-to-end learned codec with an entropy-coded latent.
wai.image.jpegai In the standard Integer entropy decode in the reference sink; the exact decoder calls no platform maths library. A variable-frame-rate neural codec for very-low-bitrate speech.
A low-frame-rate codec covering speech, music and general audio in one model.
A mel-spectrogram patch-based VQ-VAE with a HiFi-GAN vocoder; a simpler structure than stacked residual quantizers.