Opus 1.1 Released 62
New submitter rvalles writes "Xiph.org just released a major update to their Opus audio codec. Opus 1.1 offers major improvements over last year's 1.0.2 release. Opus is a general-purpose, very flexible, open and royalty-free audio codec that offers low-latency and high quality/bitrate, incorporating technology from Skype's SILK codec and Xiph.Org's CELT codec. Its first release beat everything else last year at 64kbit/s in a listening test held at HydrogenAudio."
Re:Nice, impressive achievement (Score:5, Informative)
Depends on what mobile device? The reference code has extensive ARM optimizations, that's in fact one of the main improvements in 1.1 And yes, it can be accelerated with a programmable DSP if present, IIRC there's some support for C55x in the same reference code.
Audio decoding is fast enough on modern ARM SoC that dedicated hardware isn't strictly needed to get good battery life.
Re:Oh lookie (Score:5, Informative)
AMR is pretty widely used as a voice codec, Ogg is used in most major AAA games, and as for Opus/SILK, you might have used Skype before...
Already has good adoption (Score:5, Informative)
Opus wasn't designed for audio files, but for streaming audio. In that realm it's adoption looks very promising. It has already been integrated into the Skype codebase and will likely be used in the next major release of Skype. It is also one of two mandatory audio codecs for in the draft for WebRTC, which is a new standard for browser-based chatting.
Re:Oh lookie (Score:5, Informative)
Sure, this was originally Skype, but Microsoft has continued to work with us even after acquiring Skype.
Re:Nice, impressive achievement (Score:5, Informative)
The thing about audio encode/decode is that its relatively low MIPS - with todays mobile CPUs its almost not worth the complexity to offload it to DSP. During a call your CPU has to stay awake anyway and drain battery, there is very little wattage saving moving it to DSP. It would only make sense if you are dealing with multiple, and i mean more than 2 simultaneous encoded streams ( decode is cheap ). The story was different a few years ago when the dominant CPU was ARM9 running in a 150-200 mhz range, where audio codec easily chewed up 50% or more available MIPS.
Video encoding is a whole different matter of course.
Re:Nice, impressive achievement (Score:4, Informative)
Also note - any sort of offload will add some latency, because you have to have a buffer between DSP and main CPU for them to run asynchronously. That latency is often undesireable.
Re:Nice, impressive achievement (Score:4, Informative)
You didnt get it - the speech codecs encode data at 10 millisecond or 20 millisecond intervals, depending. Sometimes 50-60 millisecond multiframe packets. For the two cores to work asynchronously, you have to hand over the minimum of one frame, for efficiency's sake preferrably more. So minimum incurred latency is at least one frame or 10 milliseconds - normally more in offloads.