STIPA Operation Guide
This is the help guide for STIPA, the speech intelligibility measurement tool. STIPA works by analyzing how much of the amplitude oscillations embedded in a special audio test signal can be extracted, to determine how easy it will be to understand someone speaking in the room, and by extension how well a sound system will be understood.
STIPA has been adopted by many local and national fire codes and other public standards as the tool of choice for measuring speech intelligibility.
Our implementation of STIPA is based on v4 of the standard and complies with all sections of the standard.
STIPA
Pro and STIPA Basic
This guide applies to both STIPA Pro and STIPA Basic. The same calculation algorithms are used for both modules. STIPA Pro includes the option to turn on an off level-dependent auditory masking, STIPA Basic does not.
STIPA Basic does not include Save / Recall, Save to Photo Roll, Test Averaging, LEQ Display, and the Detail View screen. It is suitable for field STIPA measurements, and uses the same test signals as STIPA Pro.
This section describes how to use STIPA to measure speech intelligibility.
Setup
The STIPA signal will be played in the room, and STIPA module will sample the signal using the built-in iOS device microphone, or an external mic. Results will be shown as the test is running, and at the end of the test.
You can play the STIPA signal from the App itself, or you can download it from our website and burn it to a CD. Do not alter the signal in any way and avoid compressing it (keep it as a WAV file, do not change to AAC or MP3 in iTunes). Also, if the sample rate of the playback device varies from the sample rate that the iOS device is running at by more than a few percent, it will skew the results.
Running a STIPA Test
Once you have the signal playback set up, you are ready to run a test. Note which of the 6 test result slots is selected (the highlighted segment), as this is the test number that the results will be stored in. (See Averaging, below).
Start the signal playing, and tap the RUN button. The test signal analysis will begin, and you will see a progress bar begin to move across the screen. While the test is running, the only other sound should be normal room background noise. At the end of the test (20 seconds), the test will automatically stop. You can stop the test before 20 seconds have completed, and the results of the tests up to that point will be shown, but they may not be as precise or accurate as a full 20 second test.
At the end of the test you can see the STIPA overall value (the large number on the screen), or you can inspect the individual m1 and m2 values by tapping the Details button.
While the test is running, the SPL levels for each of the octave bands, and the overall SPL level, are shown on the screen. At the completion of the test, the Leq (average SPL) for the entire test are shown, and stored with the test results, so that they can be recalled at a later time.
Averaging - Pro Feature
The STIPA standard suggests averaging six 15-second tests, and discarding or repeating any tests that vary more than 0.02 from the mean. To provide for this, we have set up our App with 6 test slots, and an average button. Select any of the 6 test slots and run a test. When complete, the test slot number will be green. This only indicates that a test has been completed for this test slot number, not the test pass/fail status. Unused test slots will appear in red. The number of completed tests will be shown below the test slot selector.
You can select test slots at any time by tapping the number of the test. To view the average of all of the completed tests, tap the ҁvgӦ#10;button.
Adding Occupational Noise to a Test Run – Pro Feature
In this case you can save a recording of the noise during the day and add it to the test that you are running at night. To do this, follow these steps:
-Record the Noise Navigate to the Recorder module, on the Utilities menu. Go to the settings page, and select Mono, 48kHz sample rate, and WAV compression (none). Turn off Record Monitor to prevent feedback. On the recording page, make sure the gain slider is set to 0dB. Now record a sample of the occupational noise in the space. You should make this recording longer than a typical STIPA test, but if it is shorter the recording will be looped during the test run. Save this recording using the Save/Recall page, and make note of the name.
- Select the Recorded Noise When you are ready to do the test, from the STIPA module, select the settings page, and tap the button under the Added Occupancy Noise label. This may say “Select Stored Audio File”, or it may be the name of a previously selected file. Pick the desired noise file from the list and tap recall. If the file sample rate is not correct you will see an error message. Tap Done to return to the main screen.
- Run the Test Run the test as usual. The recorded noise file will be silently mixed in with the live incoming audio prior to the STIPA calculations. You will see “Using External Noise File” on the screen beneath the test selector.
- Clearing the Noise File If you no
longer wish to use the recorded noise, you can either go to the settings page
and tap “Clear Noise File” or simply exit back to the menus and re-open STIPA.
Note that the noise is included in the test results, and there is no way to
remove or add noise to a test that has already been run.
Results
There
are several items shown on the screen for each test.
Overall STIPA value This value is shown in large digits on the screen. Larger numbers are better, i.e. 1.00 is “perfect”.
Tests Status Completed tests are shown in green on the test selector bar, and unused test slots are shown in red. Tap Avg to see the average of all completed tests.
Test Qualification Band One of the letter slots will be highlighted, indicating the relative speech intelligibility level. The highlight is also color-coded, with green = good or excellent, yellow = fair, and red = poor.
SPL Leq This graph shows the Leq (average SPL) of each octave band of the test. Use it as a guide to see the signal SPL level for each octave band of the test. The yellow band on the right indicates the current SPL level at all times and is not related to the completed test results. - Pro Feature
Sample Rate Variations
If you play the test signal from your iOS device, the sample rate of the signal will match what the STIPA module is expecting. However, if you play the test signal from an external source, it is possible that the sample rate of that external device will not be exactly 44100Hz, and you may not get reliable STIPA results. To test the sample rate of the external player, find or create a sine wave signal that is at a known frequency, preferably above 4000 Hz. Play that signal while you are running the FFT module in AudioTools, and turn on Peak Tracking. Set the FFT Size to 16384 or greater.
Note the frequency display for the test signal. It needs to be within 0.05%, or 2 Hz in 4000 Hz., or 4 Hz in 8000 Hz.
Save / Recall - Pro Feature
When you save a STIPA test, all completed tests (up to 6) will be stored under the test name, and will be recalled anytime that you recall the test. Also, results from all completed tests will be included in the tab-delimited XLS file that is created when you save results. Use the normal Export function on the Save / Recall screen to export this data to a computer.
Examples of STI qualification bands
and typical applications
Category |
Nominal STI Value |
Type of message |
Examples of typical uses |
Comment |
A+ |
>0.76 |
Recording studios |
Excellent intelligibility
but rarely achievable in most environments |
|
A |
0.74 |
Complex messages,
unfamiliar words |
Theatres, speech auditoria,
teleconferencing, parliaments, courts |
High speech intelligibility |
B |
0.70 |
Complex messages,
unfamiliar words |
Theatres, speech auditoria,
teleconferencing, parliaments, courts |
High speech intelligibility |
C |
0.66 |
Complex messages,
unfamiliar words |
Theatres, speech auditoria,
teleconferencing, parliaments, courts |
High speech intelligibility |
D |
0.62 |
Complex messages, familiar
words |
Lecture theatres,
classrooms, concert halls |
Good speech intelligibility |
E |
0.58 |
Complex messages, familiar
context |
Concert halls, modern
churches |
High quality PA systems |
F |
0.54 |
Complex messages, familiar
context |
PA systems in shopping
malls, public buildings offices, VA systems, cathedrals |
Good quality PA systems |
G |
0.50 |
Complex messages, familiar
context |
Shopping malls, public
buildings offices, VA systems |
Target value for VA systems |
H |
0.46 |
Simple messages, familiar
words |
VA and PA systems in
difficult acoustic environments |
Normal lower limit for VA
systems |
I |
0.42 |
Simple messages, familiar
context |
VA and PA systems in very
difficult spaces |
|
J |
0.38 |
Not suitable for PA systems |
||
U |
<0.36 |
Not suitable for PA systems |
NOTE 1 These
values should be regarded as minimum target values.
NOTE 2 Perceived intelligibility relating to each category will also depend on
the frequency response at each listening position.
NOTE 3 The STI values refer to measured values in sample listening positions or
as required by specific application standards.
1. Abstract
This
sections describes the method for processing a STIPA v4 signal through the
STIPA module of AudioTools, which maintains compliance with the 4th edition of
the STI standard (IEC 60268-16:2011).
The purpose of this section of the document is to show that once a STIPA signal
is captured in AudioTools and processed in the STIPA application, the
determination of the STI follows precisely with the guidelines set forth in IEC
60268-16.
Since it is important for the method by which STI is determined to be accurate,
careful examination of the signal path and processing of the test signal is
provided in order to determine how the STIPA application of AudioTools complies
with the standard.
This section is presented in the form of logical signal flow, where each step
of the determination of STI is documented. The processing in AudioTools is
directly compared to the specific signal-processing step in the v4 STI
standard.
2. Introduction to the Speech Transmission Index (STI)
The determination of the Speech Transmission Index (STI) was developed in order
to quantify the reduction of the modulation envelope of speech by means of
calculating the modulation transfer function (MTF) (Steeneken
and Houtgast, 1980). The STI measure has been defined
in IEC 60268-16:2011, and provides an efficient method
of evaluating speech intelligibility in both unamplified direct conversations
as well as when the speaker is amplified through a public address (PA) system.
This is due to the fact that the STI takes into account distortions due to
noise, reverberation and echoes, if present.
The calculation of the STI through the direct measurement, the technique that
AudioTools takes for calculating STI, involves replacing actual speech with an
analytical signal, which contains the spectral and temporal characteristics of
speech. By injecting the test signal at some source location Si into a
device-under-test (DUT, in most cases, an acoustical enclosure or room), and
recording the resultant signal at an arbitrary receiver location Rj, the resultant modulation reduction factor, mr can be determined. The MTF is therefore given by:
where the MTF is determined at a given modulation frequency F by the ratio of
the modulation of the response signal mr and the
modulation depth of the probe signal mt. The calculation of the MTFs are determined for speech frequencies; octave-bands from
125-8000 Hz, fi, and for one-third octave band modulation frequencies, Fj,
between 0.63-16 Hz. More details about the test signal are given in the next
sections.
3. Test measurement setup
The following equipment is used for this validation study:
4.
STIPA test signal spectrum and modulation depth
Studio Six
Digital provides a STIPA
v4-compliant signal. The signal is also available to be played back from
within the STIPA module of AudioTools.
Looking at the way that the signal is generated shows its compliance with the
STIPA v4 standard. The generator that creates the test signal is designed to
the specifications of the standard by being comprised of the following:
The signal used for the STIPA module in AudioTools fully
complies with the signal guidelines set forth in IEC 60268-16:2011 with respect
to frequency spectrum and intensity modulation frequency and depth.
Table 1: Modulation frequencies used in the STIPA test signal
5. STIPA signal acquisition time
The recommend signal acquisition time for determining STI from the STIPA signal
is between 15-20 seconds. The duration of the signal acquisition time in the
STIPA module of AudioTools is 20 seconds, which complies with this measurement
window.
6. Integrity of the test signal
A measurement of the sample rate integrity while using the measurement
system was carried out. This is to determine if the sampling rate error falls
in the window for error. The maximum allowable error rate is 0.05%. An 8000 Hz
sine wave generated in the FFT module of AudioTools was looped back through the
analog input of the iAudioInterface2. The signal was measured with the FFT
module using a 32k point FFT. The peak of the sine wave signal was measured at
8000 Hz, which is within the 0.05% frequency discrepancy window.
7. Determination of the STI using AudioTools STIPA
The following section describes the decomposition and analysis of the STIPA
signal in the STIPA module of AudioTools for the determination of the STI. This
section deals with the signal after a portion of the signal has been recorded
into the input buffer of the software module. The input starts as a full-range
STIPA signal, which has been passed trough the system
under test, and has been degraded by noise, distortion (if present) and
reverberation/reflections. A block diagram of the signal flow in AudioTools
STIPA is shown in Figure 1. The probe signal is the reference unprocessed STIPA
signal and the response is the STIPA signal measured in the environment in
which STI is to be determined.
The STIPA signal is first bandpass filtered into seven bands from 125 Hz to 8
kHz. The signal is then rectified by squaring to look at the intensity envelope
within a band. In order to extract just the low-frequency intensity
modulations, the signal is then decimated by 150x. The native sampling rate of
the STIPA module of AudioTools is 22050 Hz. The signal is first decimated by
15x, and then 10x. The sample rate after decimation is 147 Hz. The
transmission index (TI) is then determined by comparing the intensity envelope
of the probe to the response. This entire process is repeated for each of
the seven octave bands of noise and then the STI is determined by taking a
weighted sum of each of the bandճ TIs.
FIGURE 1: Block diagram for STI calculation.
a. Filter STIPA into Octave Bands
The first stage of signal processing involves filtering the STIPA signal into
seven octave bands. This is accomplished with 8th order IIR ANSI Type 0 filters
centered at 125, 250, 500, 1000, 2000, 4000 and 8000 Hz.
b. Signal Rectification
In order to extract the intensity envelope of the test signal, it must be
rectified. This is accomplished by squaring the input signal. The rectification
comes directly after band pass filtering.
c. Low-pass filter through decimation
The signal must be low-pass filtered next to extract only the low frequency
modulations that are analyzed in the calculation of the STI. This is
accomplished through a two-stage decimation filter. The audio coming into the
STIPA module of AudioTools is sampled at 22050 Hz. The decimation filters are
15x and 10x 8th order IIR ANSI Type 0 filters. They are applied to each of the
bands of filtered and rectified STIPA signal. They are applied in series. After
both decimation filters, the new sampling rate becomes 147 Hz for the remainder
of the STI calculation.
d. Determination of the Modulation Reduction Factor (MRF)
After the band-passed STIPA signal has been
decimated, the next step is to determine the ratio between the response and
probe signals of the intensity modulation at each of the two frequencies in
Table 1 for each octave band. The ratio of each is the modulation transfer
function (MTF). The MTF is then converted to a signal to noise ratio (SNR) by
dividing the MTF by 1-MTF. That quantity is multiplied by 10log10. Per the
STIPA standard, the SNR is restricted to a range of +15 dB SNR and -15 dB SNR.
If the SNR is above 15 dB SNR, the value is transformed to 15 dB, and if the
value is below -15 dB SNR, the value is transformed to -15 dB SNR. Then the SNR
is shifted by 15 dB and divided by 30 in order to generate the TI, which is a
value between 0 and 1. This is calculated for each of the modulation
frequencies in each of the seven octave bands. The TI for each octave band is
then averaged across the two modulation frequencies in the band.
e. Correction of MRF based on Level Dependent Auditory
Masking
An option in the STIPA module of AudioTools that is turned on by default, in
order to comply the with STI v4 standard, is the correction of the TI for each
octave band based on the sound pressure level (Leq)
of the octave band directly below the band being analyzed. The STI standard
uses this to account for level dependent auditory masking. AudioTools STIPA
calculates the Leq for each octave band after
filtering during the recording process. The correction takes places for the 250
Hz Р8000 Hz octave bands. AudioTools STIPA calculates the auditory masking
factor for each of the octave bands using the formulas in section A.3.2 of IEC
60268-16:2011 and multiples the calculated TI by this corrected factor.
f. Calculating the mean modulation transfer function
The final determination of the STI is completed by taking a weighted average of
each octave bandճ adjusted TI. In order to comply
with the v4 STIPA standard, the frequency-weighting
factors: αk, and redundancy-correction factors: βk are employed from Table A.3
in IEC 60268-16. These values are reproduced in Table 2 in this document. The
male weighting factors are used, since the STIPA signal has a male spectrum.
The weighted sum is determined and the resulting display of the STI is
displayed on the STIPA output screen in AudioTools. In addition to the STI
number, a nominal qualification band is given, which has been derived from the
Annex F of IEC 60268-16 standard. These nominal grades are from U to A+.
TABLE 2: Frequency-weighting factors: αk and redundancy-correction
factors: βk. Factors assume male speech and the sum of αk Рsum
of βk = 1.
8. Conclusion
The STIPA module of AudioTools has been designed following the careful guidance
of IEC 60268-16, v4. The standard provides step-by-step instruction for both
the generation of a v4-complient STIPA signal as well as the decomposition of
that signal in order to extract the necessary information to determine the STI.
The STIPA module of AudioTools performs each of the necessary steps to the
recorded STIPA signal that have been described in the current STI standard. In
addition, the quality of the filters in AudioTools are designed to exceed ANSI
Type 0 standards.
9. REFERENCES
IEC 60268-16, ғound system equipmentАart
16: Objective rating of speech intelligibility by speech transmission index,ӠIEC, Switzerland (2011).
Steeneken, H. J. M. and Houtgast,
T., ҁ physical method for measuring speech transmission qualityӬ
J Acoust Soc Am 67, 1980, 318-326.