MOS Score Calculator

Sometimes you’re in the middle of a video call and everything sounds great for a couple of seconds and then goes quiet or turns robotic. It is frustrating enough by itself. But when it happen while pitching a client or delivering an important status update to your remote team? It is a professional disaster. And if it’s happening, you know it’s not good… but without digging deep into technical specifics, which would require you to stop driving the car, so to speak, is it codec selection, packet loss, jitter, or latency?

Enter MOS: Mean Opinion Score. The idea behind MOS are to take all the subjective aspects about voice quality and try to reduce them to one number that tells network engineers and IT managers how annoying a given call will sound to a human ear. MOS scores is based off some old telecommunication studies which found out that the technical metrics don’t paint the full picture. A connection with no packet loss may still sound terrible because latency make it feel like people are talking over each other all the time.

What Is Mean Opinion Score?

So what they came up with was this sort of formula where you start with a perfect score and add penalties to factors that degrade voice quality. That way you can take into account multiple factor and the math is done for you by the calculator above (you just need to put in your environmental variables). You don’t have to remember complicated formulas or guesstimate coefficients.

The intuitive sense that missing a few packets means you’re missing a few words (or letters) makes most folks fixate heavily on packet loss statistics. That’s true, but it’s also important to consider how and when those packets is being lost. Random scattered loss hurts speech intelligibility less than burst loss. Think about having each third word dropped out of a sentence versus having one random letter dropped out of a paragraph. The first kills comprehension instantly; the second is annoying but manageable.

When setting up Quality of Service policies on your routers, this difference are critical: not all dropped packets hurt equally in ears of the listener. Many underestimate the equally important factor of codec choice. G.711 delivers perfect audio quality at practically zero processing overhead but is great for stable, wired LAN environments in which bandwidth is plentiful and cheap. There’s little wiggle room. If there is any minimal loss or jitter introduced by the network, there is no masking; G.711 will pass along these impairments to the person listening.

More moddern codecs like Opus take a small hit in theoretical max audio clarity while picking up robustness. These codecs uses advanced packet loss concealment algorithms that fill in the blanks of lost audio data based on surrounding context. The end result: a call that sounds just a bit less clear, but still conversational and understandable even during network malfunctions.

Speaking of latency: Human conversation is based on rapid turn-taking. We do this without thinking about it. Beyond 150 milliseconds one-way delay, humans tend to start cutting each other off more often. Even with good audio quality, this introduces a bit of subconscious psychological friction that causes the caller to feel like they’re being rushed, not heard.

There’s no free lunch when it comes to network engineering; the increase of your “playout buffer” setting is an attempt to help smooth out packet jitter by holding onto packets for just a little while longer before playing them out. Increasing the play-out buffer only replaces jitter with delay. Your particular use case will determine which is the lesser evil: Can you live with more glitches, or can you live with more lag?

Acoustic echo is typically caused by software configuration problems within the audio processing stack rather than faulty hardware, yet it’s easy to blame the problem on crappy hardware. The way it happens is when the speaker sound bounce back into the microphone and is heard by the remote party as a reflection of his or her own voice with a slight time delay. It’s a distraction and if you’re not careful, it could of kill a call immediately. Any serious VoIP setup requires proper echo cancellation regardless of whether the network metrics appear perfect on paper.

In the end, it’s all about balancing and striking a good MOS score. Using strong codecs and properly managing user expectations doesn’t require a fiber optic backbone to each person’s desk. Understanding how codec choices, loss, and latency affect one another will allow you to troubleshoot the actual cause instead of just throwing hardware at the problem. Begin with knowing your real world network conditions, find out which variable is negatively impacting the MOS score the most, then tweak that. Communication begins well before anyone lifts the telephone.