Report Inteligibilidad Corrected PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Acoustic Instruments and Measurements UNTREF

June, 2013

SPEECH INTELLIGIBILITY MEASUREMENT


ALEJANDRO M. ROLANDO
alerolandoush@gmail.com
Universidad Nacional de 3 de Febrero
Abstract: The propose of the current work is to measure the speech intelligibility and the background noise of
a reverberant space, in this case one position at the 2nd floor and two positions in the stairs between second
and third classroom floors. A male voice speaking some tabulated words in three different sound levels 60, 66
and 72 dB was recorded. Then, they were reproduced by a loudspeaker whose frequency response was set as
flat as possible and with the exact level to have those level values at 1m from the source. Two receivers were
placed twice on a possible listener positions in the stairs and one in the 2nd floor to obtain the speech
intelligibility in that position. The STI, C50 and NC curves are calculated. At last, some practical solutions in
order to increase the speech intelligibility are proposed.
INTRODUCTION
It is Interesting to propose the measurements of
speech intelligibility in a place that have a lot of
activity related to the speaking. The practice was
carried out in the second building of the
Universidad Nacional de 3 de Febrero located in
Caseros, Buenos Aires, Argentina. Measurements
were realized in a normal class schedule with the
people talking and talking around the area under
investigation, this helps to get a realistic speech
intelligibility of the place in a daily activity of the
university. As had been activities in the building,
the background noise is a bit high, so the speech
intelligibility is going to be affected.
EQUIPMENT
The necessary equipment to develop
measurements was:
-KRK RP8 Studio monitors (1). [SOURCE]
-DPA omnidirectional microphone (1).
[RECEIVER]
-PRESONUS Audiobox (1). [AUDIO
INTERFACE]
-SVANTEK 959 Sound Level Meter (1).
[BACKGROUND NOISE]

Alejandro M. Rolando UNTREF (2013)

the

The software to record and analyze was:


-Cubase 5 to record the signals.
-Adobe Audition with Aurora Plug In to analyze
the files.
LOCATION
As mentioned above, the location of the source
and receiver was in 3 different points.
1st position: At second floor in the main space in
front of the bathroom for ladies.
2nd position: At the stair landing of the 2nd floor
stairs.
3rd position: At the stair landing of the 1st floor
stairs.
These three zones are typical points of the
meeting where the students and professors talk
each other. In case of the stairs location, there is
not acoustical treatment and the shape makes the
place be very reverberant, therefore the spoken
words are not expected to be fully understood.
The Source Receiver points are shown in Figure
1:

Page N 1

BACKGROUND NOISE

Figure 1: Image of the 3 points of measurement.

For the 2nd and 3rd positions, the source was


located at the 2nd position and the only that was
moved was the microphone, 1,5 m from the
source (2nd position) and 2 m to the 3rd position
(Figure 2 and 3)1.

The background noise was measured in the first


measure position in order to obtain the relation
between the signal and the noise for the further
analysis of the STI feature. The instrument used to
register this value was the sound level meter
Svantek 959, with slow time integration and 30
seconds of duration.
The values are:
Leq A: 61, 5 dB.
Leq Z: 68, 9 dB.
SIGNALS
The signals used to obtain the STI parameters
were: a) Log Sine Sweep b) Recorded tabulated
words.
a) Was reproduced the log sine sweep to get
the Impulse Response of the place that
the measurement took place. The
duration of the sweep was 60 [sec].
b) A speech of male student was recorded
speaking the following tabulated words
at 1 meter from the microphone 3 times,
each with a different voice level (60, 66
and 72 [dB]).

Figure 2: 2nd measurement position.

Then, the recorded words were reproduced by the


KRK studio monitor and captured by the DPA
microphone.
SPEECH INTELLIGIBILITY INDEX

The second source (bellow) in Figure 2 is off.

There are several features that could describe the


speech intelligibility of a room as the syllable
intelligibility. It consists of a person speaking
aloud some tabulated syllables, whose vowel and
consonants are correctly distributed. The process
takes place in the room under test. Inside the room
there are some listeners who will go writing a list
of the syllables they hear. The amount of correct
syllables written well realizes if the speech
intelligibility is god or bad in this place. This

Alejandro M. Rolando UNTREF (2013)

Page N 2

Figure 3: 3rd measurement position.

technique requires a speaker and multiple


listeners; it is a primary approximation to know if
the word is going to be well understood.
Another speech intelligibility parameter that could
explain if an enclosure is proper to understand
speaking is the C50. The Clarity parameter C50
is the ratio between the energy arrived at times
before and after 50 milliseconds. It is a good
index to characterize the speech intelligibility in a
room.
C50 is expressed by:

obtained by applying 4 modulation frequencies at


500 Hz and 5 modulation frequencies at 2000 Hz.
Which STI values indicate good speech
intelligibility?
According to the IEC 60268-16 standard, the STI
values are characterized as follows:
Table 1:
STI values:
BAD
0 0.3

POOR
0.3 0.45

FAIR
0.45 0.6

GOOD
0.6 0.75

PROCEDURE
(1)

This parameter include the energy arriving at the


first 50 msec, that represent the early reflections,
separating them from the late reflections because
if take into account the reflections far from the
direct sound the speech intelligibility are going to
be bad2. When the energy overcomes the 50msec
up to 80msec the index is C80 and is
appropriate to live music in closed spaces.
What is STI?
STI means Speech Transmission Index, is a
representative parameter to evaluate the speech
intelligibility of an enclosure. To understand well
this feature, needs to know what MTF is.
The Modulation Transfer Function (MTF)
quantifies the smoothing effect of amplitude
modulation of the simulated speech signal by the
background noise and the reverberation of the
space. It varies according to the signal spectral
components and the enclosure that is analyzed,
because the decay of the signals is going to
respond to the shape, volume and coatings of it.
Hougast and Steeneken developed the concept of
STI by converting the MTF values in 7 octave
bands with 14 modulation frequencies. Also, there
is another index that emerges from STI called
RASTI (Rapid STI). This is a simpler version

It considers the late reflections to the


reverberant field, so it is better for live music
than for understand words.

Alejandro M. Rolando UNTREF (2013)

The source was installed in the first position at a


typical male height rounding the 1, 70 meters. The
microphone and sound level meter were located at
1 meter from the source in straight line and was
measured the level to set the source to reproduce
at this value. Then, the background noise was
recorded.
Once the level of the source was determined and
the background noise was measured, the
reproduction of the signals began. The first one
was the log sine sweep in order to get the impulse
response of the space. Then reproduced the speech
signals with the lowest level (60 dB). The speech
signal was repeated at 66 and 72 dB.
While all the people inside the building continue
their activities, the microphone recorded
everything: Signal + background noise.
After the measurements in the first position, the
equipments were moved to the second one with
the same arrangement except that the source and
the receiver distance was 1, 5 meters. Then, the
receiver was moved to the 3rd position about 2
meters away from the source. The distances of the
receiver were determined in order to simulate the
conversation when a person is going downstairs
and the other upstairs or if both are going up or
down3.

There is a common situation inside the building;


several people are going up/down and meet in
the stairs where they talk for few seconds.

Page N 3

EXCELLENT

0.75 - 1

ANALYSIS OF THE RESULTS

Each box has its own color that represents the


degree of speech intelligibility according to the
following scale:
Table 2:

BAD
POOR
FAIR
GOOD
EXCELENT
In the tables below can see the STI values
analyzed in the human voice frequency spectrum.
The 500 Hz to 4000 Hz frequency range are going
to be analyzed.
Table 3:
Speech Level / F [Hz]
60 dB
66 dB
72 dB

500
0.567
0.702
0.789

1st Position
1000
2000
0.556
0.606
0.824
0.813
0.894
0.868

4000
0.640
0.784
0.826

Speech Level / F [Hz]


60 dB
66 dB
72 dB

500
0.470
0.515
0.583

2nd Position
1000
2000
0.184
0.425
0.530
0.596
0.576
0.628

4000
0.557
0.662
0.684

Speech Level / F [Hz]


60 dB
66 dB
72 dB

500
0.349
0.407
0.424

3rd Position
1000
2000
0.355
0.357
0.448
0.435
0.466
0.452

4000
0.414
0.470
0.483

st

In the 1 Position, measured at 1 meter from the


source, the STI increases according to the
frequency and the level of the signal. The values
obtained are almost excellent.
In the second one, measured in the stairs at 1,5
meters from the source, most of the STI values
are fair and increases in the 4 KHz octave band at
high source levels.
In the last position (2 meters from the source),
most STI values are poor except in the case of 72
dB level source, where the values increases to be
fair from 1 KHz to 4KHz.

Alejandro M. Rolando UNTREF (2013)

The reason why the speech intelligibility


decreases as changing the positions is because in
the first place the reflections are attenuated and
the ceiling has some absorption, so the
reverberation time there is quite low. As said
above, long values of reverberation time decreases
the speech intelligibility so it is expected that have
low STI values in the stairs, where the geometry
and the surface materials on the walls make the
place to be more reverberant.
The measured reverberation time of each position
are detailed as follows:
Table 4:
Freq. [Hz]
125
250
500
1000
2000
4000
8000

RT [sec] 1st Pos. RT [sec] 2nd Pos. RT [sec] 3rd Pos.


0.497
1.430
1.397
0.702
1.308
1.235
0.704
1.605
1.693
0.691
1.887
2.109
0.785
2.023
2.143
0.772
1.741
1.965
0.713
1.408
1.590

As expected, the reverberation time increases in


the stair positions because of the excessive
reflections and no acoustical treatment. In the
second one at 2 kHz there is 2 seconds of RT and
can see that increasing the level of the source (or
speaking stronger in real situations) the speech
intelligibility is better, in the third position the
speech intelligibility is poor, even at the loudest
level of speaking.
The C50 values are shown below:
Table 5:
C50
1st Position
2nd Position
3rd Position

250
11.289
3.955
-2.693

500
12.618
1.339
-7.151

1000
14.150
2.287
-1.781

2000
12.306
2.774
-3.772

4000
10.304
3.941
-1.963

The C50 values are getting worse as changing the


positions. This serve to confirm that in the first
position the intelligibility is quite better and
worsens in the stairs location, as it had been
predicted with the reverberation times.
SPEECH INTELLIGILITY BY GENRE
Table 6:

60 dB
66 dB
72 dB

1st Position
MALE
FEMALE
0.601
0.605
0.773
0.772
0.828
0.828

Page N 4

60 dB
66 dB
72 dB

60 dB
66 dB
72 dB

2nd Position
MALE
FEMALE
0.421
0.429
0.576
0.582
0.615
0.625
3rd Position
MALE
FEMALE
0.379
0.376
0.444
0.447
0.461
0.466

In the common space (1st Position) there is almost


no difference in the speech intelligibility between
men and women. But in the 2nd and 3rd, the
FEMALE voice makes the speech intelligibility to
be better than the MALES. The female voice has
higher frequency components than male voice.
The high frequencies use to have lower RT values
than the low frequencies so, as said, better
intelligibility; in other words, lower RT values at
the dominant frequencies of the female voice
results in a better intelligibility. According to the
previous, the female voice is shows more clarity
than the males voice.
CONCLUSION
The speech intelligibility in the common spaces of
Caseros II building (Untref) is FAIR, but in the
stairs is POOR. This is not the worst situation, but
it could be fixed by applying acoustical
improvements. The acoustic treatment could be
oriented to reduce the reverberation time by
placing absorbent panels on the ceiling and walls.
The background noise is very difficult to be
controlled, and according to the NC curves for
classroom ambience (35 to 45 db), it exceeds the
stipulated noise values.
There remains the study of the speech
intelligibility
in
classrooms
where
the
qualification must be from FAIR to
EXCELLENT, due to the student cannot lose
what the teacher is speaking.

Alejandro M. Rolando UNTREF (2013)

Page N 5

You might also like