Daala

Daala
Name	Daala
Developer	Xiph.Org Foundation, Mozilla, Broadcom, Google
Released	2013
Latest release	2015 (development ceased)
Programming language	C, assembly
Operating system	Cross-platform
License	BSD

Daala is an experimental video codec project initiated by engineers at the Xiph.Org Foundation with contributions from teams at Mozilla Corporation, Google, and Broadcom Corporation. Conceived as a next-generation successor to codecs such as Theora (video codec), Daala explored novel coding tools and perceptual models to compete with contemporaries like H.264, H.265, VP8, and VP9. The project produced prototypes, design papers, and reference implementations that influenced later standards and open-source encoders.

History

Daala began within Xiph.Org Foundation research efforts to create a free, royalty-unencumbered codec alternative in the wake of projects including Ogg Vorbis and Theora (video codec). Early architecture and algorithm research involved contributors from Mozilla Corporation and hardware partners such as Broadcom Corporation and attracted academic collaborations with researchers from institutions like Queen Mary University of London and University of Cambridge. Public demonstrations and code releases appeared around 2013–2015, while active development slowed as several Daala innovations were integrated into the Alliance for Open Media initiative and the AV1 codec. The project’s repository, experimental builds, and conference presentations appeared at venues including IETF, SIGGRAPH, and IEEE ICIP.

Design goals

The primary aims included creating a royalty-free codec that delivered compression efficiency competitive with H.265 and AV1 while avoiding patent encumbrances associated with proprietary codecs from entities like MPEG LA. Daala prioritized perceptual quality using psychoacoustic and psychovisual principles studied by researchers affiliated with IRCAM and MPEG. Goals emphasized low-complexity encoder/decoder trade-offs relevant to implementers such as Broadcom Corporation for set-top boxes and Raspberry Pi-type hardware, as well as software stacks maintained by Mozilla Corporation and Google. Interoperability and open licensing were central to align with initiatives like Xiph.Org Foundation’s previous work on Ogg Vorbis.

Technical architecture

Daala’s architecture diverged from block-transform centric designs exemplified by H.264 and VP8. It employed a lapped transform approach influenced by research from JPEG 2000 and Dirac (video codec), replacing traditional block-based deblocking filters like those in H.265 with pre/post-filtering to reduce blocking artifacts. The codec used directional intra prediction strategies reminiscent of those explored in HEVC research and entropy coding schemes leveraging range coding techniques from CABAC research. Daala’s bitstream and container interactions targeted integration paths with formats such as Matroska and transport with RTP and MPEG-TS for application scenarios similar to YouTube and Vimeo streaming.

Compression techniques

Key innovations included the use of the Perceptual Vector Quantization (PVQ) algorithm, which took inspiration from encoder methods used in FLAC and vector quantizers in audio codec research at Xiph.Org Foundation. PVQ combined shape-gain quantization with perceptual masking strategies similar to approaches in AAC (file format) psychoacoustic modeling. Lapped transforms reduced ripple and blocking artifacts like those addressed by JPEG XR, while a novel deringing filter replaced post-filtering schemes used in VP9. Entropy coding leveraged range coding advances from ITU-T and ISO/IEC standard research, and motion compensation borrowed hierarchical techniques akin to those in MPEG-2 and H.264/AVC scalable extensions.

Performance and comparisons

In prototype evaluations, Daala demonstrated competitive subjective quality on certain content compared with VP9 and early HEVC encoders, particularly on low-bitrate sequences where blocking and ringing artifacts dominate. Objective metrics such as PSNR and SSIM gave mixed results relative to optimized x264 and x265 encoders, reflecting Daala’s emphasis on perceptual metrics over pixelwise fidelity. Hardware-friendly aspects appealed to implementers like Broadcom Corporation, but real-world adoption lagged because of encoder maturity and tooling compared to production encoders used by Netflix and Amazon Prime Video.

Implementations and usage

Reference implementations were written in C with assembly optimizations for architectures supported by vendors including Intel Corporation and ARM Holdings. Experimental integrations appeared in media frameworks such as FFmpeg and players like VLC media player for testing and demonstration. Although not standardized by bodies like ISO/IEC JTC 1, Daala code and ideas were referenced by contributors to the Alliance for Open Media and informed AV1 development at organizations including Google and Mozilla Corporation. Academic prototypes and test clips circulated on platforms used by the research community, including arXiv preprints and presentations at ACM Multimedia.

Legacy and influence

Daala’s lasting contributions are methodological rather than commercial: PVQ, lapped transforms, deringing strategies, and perceptual tuning informed work in AV1 and later open codecs. Engineers from Xiph.Org Foundation, Mozilla Corporation, and Google carried concepts into collaborative projects within the Alliance for Open Media and into encoder projects like aomenc and research forks hosted on GitHub. The project influenced teaching and papers at universities such as Massachusetts Institute of Technology and University of California, Berkeley, and its artifacts continue to appear in comparative codec surveys by organizations like ITU and Video Quality Experts Group.

Category:Video codecs