Microsoft has announced a change to the well-known Silk audio codec introduced by Skype for Business. The Silk solution appeared over a decade ago and offered the technology to transmit voice over the internet through VoIP. The codec used in VoIP back then was the G.722, and it required 64Kbps to transmit wideband audio. Silk transformed the landscape by delivering wideband quality with just 14kbps.
Now that we’re entering a new decade of communication, users have access to everything from 5G, to high-speed broadband. However, there are still segments of the Microsoft userbase limited to lower connection speeds. These teams experience significant packet loss and network disruption. According to Microsoft, the utilisation of available bitrate is still important today, just as it was in the dial-up world. That’s where Microsoft Satin comes in.
The Satin AI-powered audio codec can deliver so much more for the future of audio.
A Stronger Audio Codec Experience
Designed to deliver excellent audio even despite high pocket loss, Satin can provide super wideband speech with a bitrate of only 6kbps. The AI-powered codec provides excellent quality at lower bitrates, so Microsoft can use more of the bandwidth available to ensure communication resiliency.
Microsoft notes that the team built the new codec with decades of algorithmic experience and advanced techniques in machine learning. Satin is capable of redefining super wideband audio in the modern landscape, allowing for a new level of audio performance. Microsoft explained in a blog post about the new codec, that Satin uses a deep understanding of speech production, acoustics, and modelling to encode sparse representations of the signal.
Satin also only encodes and transmits specific parameters in low-frequency bands, to reduce the bitrate required even further. Through deep neural networks, Satin can estimate high band parameters at the decoder from the low band parameters received. The Microsoft Team has thoroughly tested the Satin codec and has optimised the experience to ensure a 40% reduction in computational complexity. This means that the codec can easily run on all devices.
