Last week, the team at Opus announced the general availability of Opus Audio Codec version 1.3. Opus 1.3 comes along with a new set of features, namely, a recurrent neural network, reliable speech/music detector, convenience, ambisonics support, efficient memory, compatibility with RFC 6716 and a lot more.
Opus is an open and royalty-free audio codec, which is highly useful for all audio applications, right from music streaming and storage to high-quality video-conferencing and VoIP. Six years after its standardization by the IETF, Opus is included in all major browsers and mobile operating systems, used for a wide range of applications and is the default WebRTC codec.
New features in Opus Audio Codec 1.3
Reliable speech/music detector powered by machine learning
Opus 1.3 promises a new speech/music detector. As it is based on a recurrent neural network, it is way simpler and reliable than the detector used in version 1.1.The speech/music detector in earlier versions was based on a simple (non-recurrent) neural network, followed by an HMM-based layer to combine the neural network results over time. Opus 1.3 introduces a new recurrent neuron which is the Gated Recurrent Unit (GRU).
The GRU does not just learn how to use its input and memory at a time, but it also promises to learn, how and when to update its memory. This, in turn, helps it to remember information for a longer period of time.
Mixed Content encoding gets better
Mixed content encoding, especially at bit rates below 48 kb/s, will get more convenient as the new detector helps in improving the performance of Opus. Developers will experience a great change in speech encoding at lower bit rates, both for mono and stereo.
Encode 3D audio soundtracks for VR easily
This release comes along with ambisonics support. Ambisonics can be used to encode 3D audio soundtracks for VR and 360 videos.
Opus detector won’t take much of your space
The Opus detector has just 4986 weights (that fit in less than 5 KB) and takes about 0.02% memory of CPU to run in real-time, instead of thousands of neurons and millions of weights running on a GPU.
Improvements in Security/hardening, Voice Activity Detector (VAD), and speech/music classification using an RNN are simply add-ons. The major bug fixes in this release are CELT PLC and bandwidth detection fixes.