The release a couple of months back of the Skype v4.o client for Windows was noteworthy as the introduction of their in-house developed SILK codec. Earlier today during an eComm 2009 presentation Jonathan Christensen, Skype GM Audio & Video, announced that SILK was being released under a royalty free license.
SILK was notable as being capable of narrowband (8KHz), wide band (16KHz) and super-wideband (24KHz) sample rates. Skype claims the codec dynamically adapts both sample rate and bitrate in response to variable network quality. They have published a PDF with a very general overiew of codec performance expressed in terms of bitrates, CPU requirements and MOS scores.
This release has fairly obvious implications for the Skype For Asterisk project at Digium. It clears the way for the creation of chan_silk allowing wideband calls to pass between Skype and Asterisk. The hope is that various hardware manufacturers and SIP trunking providers will adopt the codec as well, making wideband calling more broadly available.
With a considerable stable of hardware partners Skype will likely be able to see the codec implemented in hardware from companies like Belkin, Ipevo, Netgear, Philips & RTX Telecom. Note that none of these are names that you normally associate with SMB or enterprise telecom.
To sustain a wideband call the call path cannot touch the PSTN at all. That means IP-to-IP transit end-to-end. This is most easily done in corporate installations where voice is being integrated into the network anyway, perhaps on a WAN between regional offices. It’s harder to do with respect to the broader outside world. Hopefully there will come a day when IP-based peering will allow wholly IP based exchange of calls between parties, but that’s not the case today.
Bear in mind that others are playing this game as well. Polycom released their Siren7 & Siren14 codecs under a royalty free license last summer. There are starting to be quite a number of wideband capable codecs around but many are still saddled with burdensome licensing requirements. That alone could take them out of play in the face of newer royalty-free codecs.
It also pays to be very clear what “royalty free” actually means. You may not have to pay a per-user or per-end-point license fee, but in some cases there may well be an considerable ante to get access to the SDK. Some do this possibly as a source of revenue, but more likely as a means of not having to deal with every hobbyist programmer who feels like writing their own soft phone this weekend. It keeps only serious players making more than casual inquiries.
Finally, there is some debate about sample rates. Sampling audio at 8 kHz results in a usable pass-band of typically 3.4 KHz, consistent with a normal G.711 call…that is, narrowband. Wideband generally considered to be sampled at 16KHz, resulting in 8 KHz pass-band. There are those who propose super-wideband 24KHz sampling. And going even further, the latest open source CELT codec supports sample rates from 32-96KHz.
The human voice typically has limited spectral content above 10 KHz so there’s a point of diminishing return with regard to ever-higher sample rates if the application in question is telephony. There are some in-the-know who argue that the difference between wideband and super-wideband is not appreciable. Of course, if the application is conveying music then that’s a whole ‘nuther matter.
With companies like Audio Codes adopting a pro-wideband stance we may finally see hardware gateways that can handle a variety of wideband codecs. It’ll certainly be interesting to see how SILK fares in the marketplace. For now the broad hardware support is behind the more established G.722 and AMR-WB, but that could change.