My comments on intelligibility in the February issue sparked a flurry of emails and questions concerning the measurement of STI and STIPA in particular. Therefore, I thought it might be appropriate to discuss this topic a little more. One of the most-
asked questions relates to which manufacturer’s meter I use when making my own STI measurements. On the surface of it, this is a natural and very reasonable question. However, if you read on, you will see it is not one that I feel I can directly answer because I will not publicly endorse or criticize equipment. If you come to one of my talks or lectures, however, it might be reasonable to assume that I would not be using a piece of equipment that I knowingly know to be incorrect or inaccurate.
The question did, however, make me do a mental count of the various STIPA measuring instruments I have at my disposal. The answer is, perhaps, a surprising 10, from six different manufacturers! (That excludes iPhone apps, so adding these in would bring the total to 12…but I don’t really count these.).
Now, some people might find 10 STIPA meters a little excessive (my bank manager certainly did), but perhaps I should put this is perspective. Before I do, though, I should really increase the count to 10½ because I also have a B&K RaSTI meter. Some of the meters are dedicated STIPA meters, but most are sound level meters or Real Time Analyzers with a STIPA module.
But why 10 (OK, 10½)? It could be that, as chairman of the IEC standards committee for the measurement of Speech Intelligibility by means of the Speech Transmission Index (STI), I need to keep up to date with which meters are on the market and how accurate they are. The other reason could be that I like collecting meters!
Now, rather like the man with a single watch who is sure he knows the time but is then confused when he buys a second watch that reads a different time, so it can be with STIPA meters. OK, so that might explain three meters; what about the other seven? Well, in three cases, I have two different audio/sound-level meter analyzers from the same manufacturer (older and newer models). Two I bought to test because I was suspicious of the results they were producing, and two I acquired as a part of a test package.
So what is STIPA anyway, and why all the fuss? STIPA stands for STI for PA systems and was developed around 2000 to provide an easy means of measuring the potential speech intelligibility of public address and voice alarm/emergency sound systems. Over the years, it has become the accepted international standard for measuring PA systems. It was formally introduced to the world in 2001 in an acoustics conference paper and was internationally standardized in IEC 60268-16 published in 2003. This was the third edition of the STI standard. The standard was updated and expanded in 2011 with the publication of Edition 4 and, sometime within the next four to five years, Edition 5 will probably appear.
In order to enable STI measurements to be made using a handheld device, the complex calculations and signals involved had to be simplified. Full STI employs 14 modulation frequencies at each of seven octave band carrier frequencies, so a complex calculation based on 7×14=98 data points has to be undertaken. Although not difficult to do today, we did not have that processing power 15 years ago.
Furthermore, and more to the point, it simply is not possible to generate a seven-
octave band signal with 98 simultaneous modulations. So, instead, STIPA was formulated and uses a sparse matrix of just 14 data points. The correlation with Full STI measurements, however, is remarkably good…when implemented correctly.
That is why I have 10 STIPA meters, because not all of the manufactures have implemented the STIPA measurement technique correctly. Thus, different meters produce different answers. Over the years, I have written several AES papers on this, so will spare you the details here (they are available via the AES electronic library).
Over the past 12 to 15 years, the convergence has improved, as my jumping up and down and pointing out the errors of their ways to some of the manufacturers seems to have borne fruit. In some cases this fell on deaf ears, though, so some meters are still giving the wrong or misleading answers.
Interestingly, it is the Europeans who seem to have gotten it right with two out of my three American-manufactured meters being adrift (In real terms, this means three out of five actual meters). In some cases, the manufacturers have not kept up with the changes to the standard (in 2003 and 2011, in one case) and in others, have implemented their own interpretation that is incompatible with everybody else on the planet.
To put this in perspective, in Table 1, I set out a set of readings I made for two sets of identical laboratory conditions. (To date, each meter from the same manufacturer employs the same algorithm, so I won’t bore you with all 10 sets of data). To save gross embarrassment and to make an interesting international point, I have identified the meters as US and European.
As you can see, US 1 agrees reasonably well with the European meters, whereas US 2 is unacceptably off—particularly for test 1—and US 3 is way off.
Let me put this into context. A reading of 0.48 STI (US 2) indicates that speech will be fairly intelligible, particularly if you concentrate a little, but just fails to meet the internally recognized value of 0.50 STI required for emergency sound systems. US 3, however, rates this as being very good intelligibility. Worryingly, as can be seen by looking at the three European meters and US 1, the correct value is clearly 0.40 to 0.41, which means that only the occasional word will be understood properly and so the system under tests would be of little or no use in an emergency situation. In other words, two of the US meters are potentially putting people’s lives at risk!
Now, I do not think for one minute that this is deliberate; it’s just ignorance, and the inability of the manufacturers to follow the standard and to keep up with the changes that have occurred over the past 14 years. This is dangerous amateurism that has no place in the professional audio or life safety worlds.
So, what if it all goes wrong and you end up in a court case being sued over a system that you measured to be compliant when, in reality, it was way off? An interesting legal situation would no doubt occur, with the consultant/contractor blaming the meter manufacturer and countersuing that company. What a horrendous situation to be in, all because someone could not be bothered to either read or keep up with IEC 60268-16, the international STI standard!
Is it not reasonable to expect a non-expert consultant or contractor to know that his meter is reading incorrectly? Interesting point! It is, however, not at all unreasonable to expect that the person making the STI measurements has read (and understood) the standard and is familiar with the relevant parts of it. Thus, they should automatically be asking the manufacturers if their meters are Edition 4 compliant, and if not, why not?
Indeed, the meter display should clearly state the edition (or date) of the standard being employed. Figure 1, for example, does this very thing, and so clearly removes any such ambiguity. I would prefer that money be spent on audio systems or better measurement equipment than lining lawyers’ pockets. Don’t you?