Why Your Network Is Good Enough for Everything Except Phone Calls

VoIP From the Ground Up: Part 4 of 10

Not all data is created equal. A network can drop, delay, or reorder packets carrying a webpage and you will never notice. Do the same thing to a voice call and the conversation falls apart. This post explains why.

The real-time problem

Most internet traffic is not time-sensitive. When you click a link, your browser sends a request. The server responds. If the response takes 200 milliseconds or 400 milliseconds, you probably cannot tell the difference. If a few packets need to be resent because they were lost along the way, the total delay might increase slightly, but the page still loads correctly.

Voice does not work this way.

A phone conversation is a continuous, real-time, two-way exchange. When you say a word, the other person expects to hear it almost immediately. Not in one second. Not in half a second. The human brain is remarkably sensitive to conversational timing. Research has consistently shown that delays beyond about 150 milliseconds start to feel unnatural, and delays beyond 300 milliseconds make normal conversation nearly impossible because both speakers start talking over each other.

This means voice traffic has a deadline. Every piece of audio has to arrive within a narrow time window or it is worthless. There is no "try again later." The moment has passed.

The three enemies of voice quality

Packet-based networks introduce three specific problems for real-time audio. We touched on these briefly in earlier posts, but now it is time to name them clearly.

Latency

Latency is the time it takes for a packet to travel from your device to the other person's device. It is usually measured in milliseconds.

Some amount of latency is unavoidable. The data has to physically travel through cables, get processed by routers, and cross potentially thousands of miles of infrastructure. Light in a fiber optic cable crosses the continental United States in roughly 21 milliseconds. Add in the processing time at each hop and the encoding and decoding of audio at each end, and you are already at 50 to 80 milliseconds even under ideal conditions.

The problem comes when latency gets high enough to disrupt the natural back-and-forth of conversation. At 150 milliseconds one way, you start noticing a slight delay. At 250 milliseconds, you are stepping on each other's words. At 400 milliseconds, it feels like talking on a satellite phone from the 1990s.

Latency itself does not make audio sound bad. It makes the conversation feel bad. The audio might be crystal clear, but the timing is wrong, and that breaks communication just as effectively as static or distortion.

Jitter

Jitter is the variation in latency from one packet to the next. For a deeper look at jitter and its practical impact, see our guide on what jitter is and why it matters.

Imagine packets arriving at the destination. In a perfect world, they show up at perfectly regular intervals: one every 20 milliseconds, like clockwork. In reality, some arrive after 15 milliseconds, some after 25, some after 40. Network congestion, routing changes, and competition with other traffic all cause this variation.

The receiving device uses something called a jitter buffer to smooth this out. It collects incoming packets and holds them briefly to put them back in proper order and spacing before playing the audio. A larger buffer can absorb more variation, but it adds latency (because packets are sitting in the buffer waiting). A smaller buffer reduces delay but cannot handle as much variation, leading to gaps in the audio when packets arrive too late.

Jitter is what makes voices sound choppy or robotic. The audio is arriving, but it is arriving in an uneven rhythm that the receiving device cannot fully smooth out.

Packet loss

Packet loss is exactly what it sounds like: packets that never arrive at all.

As we discussed in the previous post, VoIP uses a delivery method that does not resend lost packets. If a packet carrying 20 milliseconds of your voice disappears somewhere in the network, that piece of audio is just gone. The receiving device has to either play silence in its place or try to guess what the missing audio sounded like based on the packets that did arrive.

Modern audio codecs are surprisingly good at concealing small amounts of loss. At less than 1%, most listeners will not notice anything wrong. But as loss increases, the concealment breaks down. At 3 to 5%, you hear obvious gaps, words that cut in and out, and a general sense that pieces of the conversation are missing. Above 5%, the call is effectively unusable.

Why other traffic does not have these problems

Web traffic, email, and file transfers all use the reliable delivery method we discussed in the previous post. Lost packets get resent. Out-of-order packets get sorted. The application only sees the complete, correct data, even if the underlying network had all kinds of problems delivering it.

This works because those applications are not time-sensitive. If it takes an extra 100 milliseconds to resend a lost packet and complete your file download, nobody cares. The file is still correct when it arrives.

Voice cannot use this approach because the resend would arrive too late to be useful. So voice traffic is exposed to every imperfection in the network in a way that other traffic simply is not. The network problems are still there for web browsing and email. They are just hidden by the resend mechanism.

The human factor

There is one more element that makes voice uniquely demanding: the listener.

Humans are extraordinarily sensitive to audio anomalies in speech. We evolved to process spoken language with incredible precision. We can detect gaps of just a few milliseconds in conversational audio. We instinctively know when timing is off. We notice tonal shifts, brief silences, and unnatural rhythms that we would never detect in, say, a music stream.

This means that voice quality issues that would be completely invisible in other applications become immediately obvious and irritating in a phone call. A 50 millisecond gap in a video stream is invisible. A 50 millisecond gap in a phone conversation is a speaker cutting out mid-word.

The bar for "good enough" is simply higher for voice than for almost anything else on the network. A VoIP quality test measures the specific metrics that matter for this bar -- latency, jitter, and packet loss -- and the MOS Explorer shows how those metrics translate into the perceived quality score that reflects what listeners actually hear.

What comes next

Now that we understand why voice is so demanding, the next post puts it all together and explains what VoIP actually is, step by step, from the moment you speak into a microphone to the moment the other person hears your words.

This is Part 4 of the VoIP From the Ground Up series. Part 5: What Is VoIP, Really? walks through the full process of how a voice call works over the internet.