Technology sharing | quick intercom, global intercom

Digital intercom is to speak voice signals through digital coding , Then it is transmitted through the Internet , The receiver decodes and plays . Voice coding in the whole intercom process , Network transmission , The playback side cache and other technical links will affect the quality of the intercom , Especially the intercom between transnational and cross operators , Large network transmission delay , High volatility , How to guarantee the intercom experience is a problem . Quick intercom is through the layer by layer optimization of technical links , The global delay is less than 300ms High fidelity digital voice intercom , How can this be done , Let's introduce one by one .

Voice is transmitted over the Internet , Have to face a problem ： Packet loss , Network fluctuations . What is packet loss , There's a little bit of explanation here , We regard the Internet as a transmission belt , So audio data is an object . Packet loss means that some objects cannot reach their destination within the effective time , You may never reach your destination . If five of our 100 items fail to reach our destination , Our transmission mission “ Packet loss rate ” Namely 5%. Yes , Internet transmission is the same , It's not 100% reliable , There is always data that cannot be transmitted to the destination on time , Or there may be confusion .

So here comes the question , If the network fluctuates , High packet loss rate , What are the consequences ？

If packet loss is serious , Wechat phone 、 Video chat or voice intercom , The voice you hear from the other person may be of poor quality , Even on and off , The call delay is very large , Or no sound at all . This can happen in any network . Whether you pass Wi-Fi、4G still 5G, Because of entering the underground garage 、 Enter the elevator 、 Poor signal coverage 、 Network bandwidth is limited , Encounter these real-time audio experience problems .

One ． Codec redundancy and packet loss resistance

The codec used for quick intercom has been specially optimized for voice , His anti packet loss strategy is different from the traditional method . In terms of communication principle , Source coding is to pursue high compression ratio as much as possible , To redundancy . Channel coding pursues strong error correction , By adding redundancy to achieve error correction . And our optimization for human voice signal characteristics is to combine the plus redundancy and the minus redundancy , Reduce redundancy where it is not important , Add redundancy in important places .

During transmission , It will split a packet into two for transmission , If the opposite end receives one of them , Then a limited distortion signal is recovered by decoding ; If the opposite end receives two packets , Then a high-quality signal can be recovered by decoding . That is, there is no need to wait for the statistics of the current network packet loss status , It is only necessary to directly implement the anti packet loss into the codec .

There are three advantages ：

1. The coding and transmission rate can be reduced ;

2. Higher quality can be achieved , The average codec level of quality achieved when a packet is received , Two packets received reach high quality codec level ;

3. Lower end-to-end delay can be realized .

Two ． Network transmission FEC

1,FEC It can be divided into two categories ： Source based and channel based .

A, Source FEC yes , The package can be delivered several times , For audio, one second can send 25 A package , Source FEC Just double 50 A package , One more time for the same size , To achieve packet loss resistance .

B, channel FEC yes , For example, the current packet loss rate 25%, We can add 50% Anti packet loss . So the original has 10 A package , After processing, generate 15 A package , this 15 Packets arrive at any 10 A package , Can achieve accurate decoding .

2, Source FEC in , If the method of multi contracting is adopted , There will be new problems , For example, the data to be transmitted is 24kpbs The voice of , When the packet is lost , It's hair 48kpbs The voice of , Two 24kpbs Send them all . Or split it into two 12kpbs Reissue ？ Each have advantages and disadvantages . If you use two 12kpbs, Reduced sound quality in exchange for packet loss resistance . If you choose 48kpbs, Maintain sound quality , Before 24kpbs If the next network packet loss is 10%, Bandwidth becomes 48kpbs after , Packet loss is also different . therefore ,Opus and Silk A new method is proposed for the encoder , The method of reducing bit rate is adopted , It's similar to two 12kbps. stay 24kbps In the audio stream of , Yes 6kbps To compensate for the previous frame . Once the big bag is lost , Just use a small package for recovery , But the problem is that the audio quality is degraded .FEC It is a good method to resist packet loss , But the problem is that it wastes bandwidth . Use FEC after , It can really improve the packet arrival rate , It can improve the quality of communication with limited delay .

FEC The flow is like this ： Three packages were sent first , from device1 Issued 3 Packages to device2,packet2 lost , Then the packet loss rate is 33%.device2 Will send a Loss info to device1, Notify it that the packet loss rate is 33%. then ,device1 Then send a new bag , A message will be sent 2 times , Two packet4, Two packet5.packet4 Packet loss , Will be another packet4 Compensation is back , such packet4 No need to resend .

From this process, we can see that , Whatever the source FEC Or channel FEC It's very important , Although a bit of bandwidth is wasted , But in the face of high packet loss , High latency environment , Ensure low delay and high quality of audio .

3、 ... and ．GT-RTN The Internet

Talk quickly with the help of anyRTC Provide us with a global based GT-RTN The Internet , In terms of backbone network transmission , Communicate through intelligent network 、 Routing planning and scheduling 、 Multiple line redundancy, etc , Widely deployed in combination with key nodes , It effectively solves the problem of network congestion 、 Transmission quality degradation caused by operator failure .

GT-RTN Through the Internet “ cloud ” Service and “ End ” Efficient end-to-end collaborative optimization , Make the unpredictable Internet manageable 、 Predictable 、 Schedulable . Based on this ability , Full link acceleration can provide users in all countries and regions with a high level of QoS guarantee .

GT-RTN The Internet has QoS guarantee 、 Full link acceleration 、 High availability 、 Four core advantages of global coverage ：

1、QoS guarantee ： Cut in 260ms The packet arrival rate exceeds 99.9%, Improve end-user access performance on average 100%.

2、 Full link acceleration ： Full link “ End ”+“ cloud ” Collaborative acceleration , Provide better overall acceleration performance and stability .

3、 High availability ： Through redundant resources and network wide intelligent scheduling algorithm , Not relying on a single physical resource , Eliminate unavailability caused by physical line failure .

4、 Global coverage ： Can be global 100 Users in multiple countries and regions provide anytime, anywhere services 、 Stable and high-speed network services

meanwhile GT-RTN Network in data security compliance , Support... During transmission TLS agreement , The communication between the terminal and the source station is in the dedicated channel , The source side can not be exposed to the Internet at all , Ensure the business security of the source station . The access of the terminal will also provide dynamic services Token Authentication and user authentication .

Four ． summary

It looks like a simple voice intercom technology , There are many technical points involved , This paper introduces how to ensure the high quality and low delay of voice in the global intercom , With these solid technologies as a guarantee, we can bring users a better use experience .