Troubleshooting Voice Quality

Nick Fox How-to, In More Depth, Lab, Network Performance

When a quick dash for the MOS is not enough

The Task

Sometimes that quick dash to fill a spreadsheet with Mean Opinion Scores (MOS) just won't cut it.

OK, we get it, your manager wants a report on the voice quality performance of widget X, so you reach for the test system.  And MOS is great, because it figures out all the factors and just gives you a single overall score, right?  So you set the system to make 100 measurements, because you want to be confident that everything is stable and, well let's face it, an average over 100 will look better than an average over 10.  Great!

The Outcome

Now here come the scores, looking good, looking good... time to grab a quick coffee... but wait - a score of only 2.6?  Now it's ok again.  Where did that come from?  And why?  Now what do you do?  There is always a choice...

What Next?

  1. Run another 900 measurements and hope the average score looks good enough?
  2. Re-write history - erase that low score?
  3. Ask a colleague if she can run this test for you, as you are too busy?

Don't worry, we're only kidding, we are professionals too!  You need to know what lies behind the low score, so here's the real plan:

  1. Bring up the 2.6 score and listen to what that voice sample actually sounds like.
  2. Flick though the diagnostic displays and look for anomalies.
  3. Identify the probable cause and add it to the report.
  4. Save the analytical results for the developers to look at using the bundled viewer app.
  5. Now get that coffee, job done and you deserve it!

That's it, now you have your coffee and a few minutes to relax, so take a look at the video for some inspiration on the troubleshooting process.

Finally you have covered the bases and your boss will love that report!

Follow-up

More on MOS?  Try these:

Video 1: Know everything about MOS - part 1: Which MOS?

Video 2: Know everything about MOS - part 2: POLQA and PESQ

Video 3: Know everything about MOS - part 3: Reference Speech Files