Opus is a lossy audio coding format developed by the Xiph.Org Foundation and standardized by the Internet Engineering Task Force, designed to efficiently code speech and general audio in a single format, while remaining low-latency enough for real-time interactive communication and low-complexity enough for low-end embedded processors. Opus replaces both Vorbis and Speex for new applications, and several blind listening tests have ranked it higher-quality than any other standard audio format at any given bitrate until transparency is reached, including MP3, AAC, and HE-AAC.
Opus combines the speech-oriented LPC-based SILK algorithm and the lower-latency MDCT-based CELT algorithm, switching between or combining them as needed for maximal efficiency. Bitrate, audio bandwidth, complexity, and algorithm can all be adjusted seamlessly in each frame. Opus has the low algorithmic delay (26.5 ms by default) necessary for use as part of a real-time communication link, networked music performances, and live lip sync; by trading-off quality or bitrate, the delay can be reduced down to 5 ms. Its delay is exceptionally low compared to competing codecs, which require well over 100 ms, yet Opus performs very competitively with these formats in terms of quality per bitrate.
As an open format standardized through RFC 6716, a reference implementation called libopus is available under the New BSD License. The reference has both fixed-point and floating-point optimizations for low- and high-end devices, with SIMD optimizations on platforms that support them. All known software patents that cover Opus are licensed under royalty-free terms. Opus is widely used as the voice-over-IP (VoIP) codec in applications such as WhatsApp and the PlayStation 4.
Opus supports constant and variable bitrate encoding from 6 kbit/s to 510 kbit/s, frame sizes from 2.5 ms to 60 ms, and five sampling rates from 8 kHz (with 4 kHz bandwidth) to 48 kHz (with 20 kHz bandwidth, the human hearing range). An Opus stream can support up to 255 audio channels, and it allows channel coupling between channels in groups of two using mid-side coding.
Opus has very short latency (26.5 ms using the default 20 ms frames and default application setting), which makes it suitable for real-time applications such as telephony, Voice over IP and videoconferencing; research by Xiph led to the CELT codec, which allows the highest quality while maintaining low delay. In any Opus stream, the bitrate, bandwidth, and delay can be continually varied without introducing any distortion or discontinuity; even mixing packets from different streams will cause a smooth change, rather than the distortion common in other codecs. Unlike Vorbis, Opus does not require large codebooks for each individual file, making it more efficient for short clips of audio and more resilient.
As an open standard, the algorithms are openly documented, and a reference implementation (including the source code) is published. Broadcom and the Xiph.Org Foundation own software patents on some of the CELT algorithms, and Skype Technologies/Microsoft own some on the SILK algorithms; each offers a royalty-free perpetual for use with Opus, reserving only the right to make use of their patents to defend against infringement suits of third parties. Qualcomm, Huawei, France Telecom, and Ericsson have claimed that their patents may apply, which Xiph's legal counsel denies, and none have pursued any legal action. The Opus license automatically and retroactively terminates for any entity that attempts to file a patent suit.
Spectrogram of Opus-encoded audio as bitrate rises (~32 to ~160 kbit/s) clearly shows lowpass behavior and better preservation of the band energy with CELT (compare original, Vorbis, MP3, AAC).
The Opus format is based on a combination of the full-bandwidth CELT format and the speech-oriented SILK format, both heavily modified: CELT is based on the modified discrete cosine transform (MDCT) that most music codecs use, using CELP techniques in the frequency domain for better prediction, while SILK uses linear predictive coding (LPC) and an optional Long-Term Prediction filter to model speech. In Opus, both were modified to support more frame sizes, as well as further algorithmic improvements and integration, such as using CELT's range encoder for both types. To minimize overhead at low bitrates, if latency is not as pressing, SILK has support for packing multiple 20 ms frames together, sharing context and headers; SILK also allows Low Bit-Rate Redundancy (LBRR) frames, allowing low-quality packet loss recovery. CELT includes both spectral replication and noise generation, similar to AAC's SBR and PNS, and can further save bits by filtering out all harmonics of tonal sounds entirely, then replicating them in the decoder. Better tone detection is an ongoing project to improve quality.
The format has three different modes: speech, hybrid, and CELT. When compressing speech, SILK is used for audio frequencies up to 8 kHz. If wider bandwidth is desired, a hybrid mode uses CELT to encode the frequency range above 8 kHz. The third mode is pure-CELT, designed for general audio. SILK is inherently VBR and cannot hit a bitrate target, while CELT can always be encoded to any specific number of bytes, enabling hybrid and CELT mode when CBR is required.
SILK supports frame sizes of 10, 20, 40 and 60 ms. CELT supports frame sizes of 2.5, 5, 10 and 20 ms. Thus, hybrid mode only supports frame sizes of 10 and 20 ms; frames shorter than 10 ms will always use CELT mode. A typical Opus packet contains a single frame, but packets of up to 120 ms are produced by combining multiple frames per packet. Opus can transparently switch between modes, frame sizes, bandwidths, and channel counts on a per-packet basis, although specific applications may choose to limit this.
The reference implementation is written in C and compiles on hardware architectures with or without a floating-point unit, although floating-point is currently required for audio bandwidth detection (dynamic switching between SILK, CELT, and hybrid encoding) and most speed optimizations.