AFMG Network Forum

mrovira · Member

Hi all,

I'm using EASE JUNIOR ver. 4.3.9.75. I usually use it to calculate few parameters (SPL Direct, SPL Total & STI) of PA systems. In EASE 4.2 or 4.1 (I don't remember) I used to use RaSTI as an intelligibility index, with noised turned on option checked and in the tutorial there were the conversion tables:

RaSTI (Short Form Formula)
0.75 to 1 - Excellent
0.6 to 0.75 - Good
0.45 to 0.6 - Fair
0.3 to 0.45 - Poor
0 to 0.3 - Unacceptable

RaSTI long Form (Peutz LF) Formula
0.6 to 1 - Very Good
0.45 to 0.6 - Good
0.3 to 0.45 - Poor
0 to 0.3 - Unacceptable

Everything was clear. Now things have improved but are more complicated.

Now there's an options tab in the calculation window where I can choose between several STI calculations (Standard, Male, Femaile, STIPa Male, STIPa Female, RaSTI).

Q1: Which one would be the best to compare PA solutions?

There are 2 check boxes,
Consider Noise Levels (which I guess it's like the old Noise turned on)
Consider Signal Masking (which I know it's explained in the standard 60268-16 but I don't have it so I have no idea of what it is)

Q2: Do I have to check "Consider Signal Masking" to get fine results analising PA systems?

Finally, now I'm using STI because is the one I've got a table (In fact I found two of it in the help file which are very similar):

STI Score Intelligibility

0.75 to 1 equals Excellent
0.6 to 0.75 equals Very Good
0.45 to 0.6 equals Good
0.3 to 0.45 equals Fair
0.25 to 0.3 equals Poor
0 to 0.25 equals Unacceptable

Syllable intelligibility STI-value

poor - 0 to 0.3
satisfactory - 0.3 to 0.45
good - 0.45 to 0.6
very good - 0.6 to 0.75
excellent - 0.75 to 1.0

But with the articles I've got like "STANDARDISATION OF PERFORMANCE CRITERIA AND ASSESSMENTS METHODS FOR SPEECH COMMUNICATION Herman J.M. Steeneken" and others, I always find:

Excel. > 0,75
Good 0,60 - 0,75
Fair 0,45 - 0,60
Poor 0,30 - 0,45
Bad < 0,30

Notice that what is considered "poor" in one table, in the other is "bad", what is "fair" or "satisfactory" in the ones I have is "poor", and so on... Including wikipedia that is where my clients first look when they want to know what is STI... and as you may understand it's not the same fair than good for the one who pays...

Q3: Where the EASE help tables come from? the 60268-16?
Q4: Is the same table valid for evaluating either STI, STIPA, RaSTI?

With the new calculation method is more dificult achieving values of 0.65 or more with an L-R PA system.

Q5: Which are the factors that are influencing the final intelligibility value in Ease Junior? Background Noise, Reverberation, ...

I'm sorry about writting this long post, but I've got lots of doubts as you see...

Thanks in advance to anyone who may help me.

Lindsay Smith · Member

You have raised many of the same questions I have.

The documentation is terrible but here is what I have found/assumed. Some of this is simplified; refer to Steeneken et al if you’re feeling masochistic.

STIPa is a subset of STI. STI divides the audio into 7 octave bands and modulates each band with 14 bands of audio. The results are combined proportionally by how much each frequency band contributes to intelligibility. STIPa also divides the audio into 7 bands but only modulates each band with two frequencies. The results are combined similar to STI except there are two sets of weighting criteria: male and female. Because males typically have lower pitched voices than females, the STIPa value for them is often lower.

I have not been able to find out what the “male” and “female” settings are for. I assume they apply the respective weighting to the full STI calcs but that’s just a guess. “Consider Noise Level” is the background noise, although I can’t imagine how an STI measurement that didn’t consider this would be at all useful. “Consider Signal Masking” adjusts for the ears diminished acuity at high volumes.

Personally for evaluating distributed sound systems I have been using STIPa male with Consider Noise Level checked. The critical thing for me is determining the speaker taps for the speakers. I assume a 10 dB crest factor and then map the total SPL, broadband, A weighted. If I like the level, then I map the STI values.

Any better methodologies, obvious problems, etc?

Lindsay

Stefan · Member

Hi,

first of all a few general statements:
1. Please make sure you are using EASE 4.3 and you are referring to the latest helpfiles, particularly Appendix A, section 6.2.2.7 which gives an overview over the STI calculation.
2. Note that EASE 4.3 is fully compliant with IEC 60268-16 r2003 which defines the STI, STIPA, RASTI and related corrections such as for male/female speakers, signal masking and noise masking.
3. In contrast older versions of EASE 4.x relied on the application of the Farrel-Becker equation which was used backward to derive RASTI from ALCONS.
4. The calculation of STI in EASE Jr is just a rough approximation. In acoustically difficult environments always the ray-tracing module AURA should be used, as the statistic (Eyring/Sabine) approach cannot account for room geometry, placement of wall materials or echoes.
5. If your business relies on STI readings please consult the IEC standard. As you may understand we cannot repeat and explain it in full detail in the EASE manual.

Now let me give a bit of an overview.

1. STI vs STIPA
Generally STI is preferred over STIPA and especially over the outdated RASTI. That is because STI utilizes all 98 modulation transfer indices, but STIPA only a subset and RASTI only a small subset limited to 500 Hz and 2kHz. STIPA has practical meaning because it seems to be sufficiently close to the STI but at the same time requires less computational power for simple handheld devices that utilize a DSP-implementation.

To be even more precise, EASE, EASERA and SysTune calculate the STI based on the IR (or its approximation) whereas typical STIPA devices use a dedicated test signal and derive STIPA results directly by measuring the modulation depth. As a result both methods are not exactly comparable. However for most practical applications the differences are insignificant.

2. Weightings
STI and STIPA can be weighted to account for the differences in the spectrum of a male and a female speaker. The male weighting is commonly used, whereas the female weighting and the standard weighting are less important.

3. Signal Masking
The limitation of the STI due to very low signal levels caused by the hearing thresholds and due to very high signal levels caused by the masking of transmitted information by neighbor octave bands can be accounted for by enabling switch for signal masking.

4. Noise Masking
The limitation of the STI due to low signal-to-noise ratio can be accounted for by enabling the switch for noise masking. In this case, the noise levels are included in the STI calculation. Note that the STI is not only a matter of S/N. It particularly considers the time structure of the system response.

5. Influence of Masking Effects
One should be aware of the practical implications of the above corrections. In many situations a minimum S/N ratio is required. At the same time a minimum STI is required too. However, if the noise floor is too high, a signal level increase may cause a decrease of the STI due to signal masking rather than an increase of the STI due to the improved S/N. This is much discussed in the community as establishments making such requirements are often not aware of their implications.

6. Criticisms
In our workshops we usually emphasize that the evaluation of speech intelligibility by means of the STI as defined in IEC 60268-16 r2003 is a great step forward compared to energy-based criteria.

However, there are fundamental effects that are not accounted for by the standard and must be recognized by the practitioner:
- The STI assessment does not take into account the quality of the frequency response. Listening tests have shown that a well-behaved system can provide much higher intelligibility ratings than one with colorations as well as dips and peaks in the frequency response.
- The STI assessment does not take into account that the human hears binaurally, that is, with two ears and that his brain applies significant processing to improve the S/N of the transmission.
- The STI assessment is quite sensitive to strong echoes. It may show significant variations based on the exact time location of a strong reflection although subjectively the intelligibility is unchanged.
- In r2003 the correction factors for the signal masking are defined by a stepped function. Small changes in the signal level may result in large changes of the STI when being close to such a step. Revision r2010 is going to define a smooth masking function.

7. STI values
STI results for one or multiple locations should always be given with a +/- uncertainty. It can be derived for example by making multiple measurements and then taking the standard deviation.
Modeling results with random-based ray-tracing methods such as AURA should also be treated carefully. It is always recommended to perform a second calculation with the same parameters. If the results do not change anymore, the resolution and particle number was high enough.

As a conclusion we should underline that the STI can be only one more indicator. It is quite valuable but it cannot replace actual measurements and listening on-site and it should always be used in combination with other acoustic measures qualifying the venue.

Stefan

mrovira · Member

Thanks Lindsay for your answer, I had seen your previous posts but I couldn't find my answers...

Thanks Stefan for this complete and extended answer, is that kind of post you have to read 3 times at least. This answers mosts of my doubts and will be very helpful for me and others with similar questions...

Stefan · Member

We are asked frequently how the STI is computed in the case of statistics-based calculations (Eyring/Sabine), when there is no full-length impulse response available.

We use the same approach as published by Steeneken, Houtgast and Plomp in their landmark article from 1980: Predicting Speech Intelligibility in Rooms from the Modulation Transfer Function, ACUSTICA, Vol. 46, 1980.

In part 6 the contribution of the direct field is discussed in relationship to the reverberant field. The squared pulse response is given by eq (11): r(t) = r_d(t) + r_r(t). This relationship can then be readily evaluated using the standard definition of the modulation transfer function, eq (A9).

We have expanded this by substituting the single direct contribution r_d by the individual loudspeaker impulses according to their arrival time, directionality and spectrum. The reverberant part r_r is based on the reverberation time and on the time of the reverberation onset (see Probe/Details).

Any practitioner of STI in EASE is recommended to take a look at the standard IEC 60268-16 as well as at this paper.

Lindsay Smith · Member

If you don't have handy access to the Acustica articles, a few papers co-authored by Steeneken are available for free download from:

http://www.steeneken.nl/sti.html

Lindsay Smith
Seattle