Using RASTI to Determine Speech Privacy

£ . „ • ... ,. .. •,... • .': .":•,'•'." •" '". ^"N tion will be very low, equal or less than if X; > + 15 dB then X; = +15 dB l - l ^ i P ^ i Tran,smissifn Analyzer t h e modulation of the background if X{ < - 1 5 dB then X; = - 1 5 dB ^ ^ ' ^ ^ ^ ^ ^ ^ ^ ^ noise. However, this measurement the intelligibility of the spoken word. , , . ' . , , . , ", m, , , . „ , „ - r

i y S s , i a | i i i t i 0m: Brief describes how problem can be solved by using elevat- The arithmetic mean of these 9 X{ val-the Type 3361 can biased to measure ed signal levels. ues is obtained and normalized to yield the Opposite speech unintelligibility, or an index which ranges from 0 to 1. speech privacy. This application can be The modulation transfer function _ used in open-plan office space to reduce c a n b e expressed as RASTI value = (X; + 15)/30 the interruptions and decreased effi­ciency caused by unintentional eaves- -, -,

■ 'f'm m m r rmr r r r r ^ ^ ^ r # ■ # # # * # # ' ♦ ' '■ ' ', L L ■ ■ # # * ■ ■ ■ m * \ * t *tr ■ ■ ■ ■ ■ ■ ■ ■ W ' ' ' ' « « i i { mV " 1 1 1 1 1 1 mA mm ■ " i * " 1 1 1 " 1 " i 1 rt " 1 * ' "* ■ * " I I ! ! ! ! , " I I

dropping; another application is for se- m(f) = — x g——

r r i t y purposes- l l i P M2*-m? 1 + 1 ° ~ ¥ " Measurements T n t r o H n r t i n n where the first term is the reverbera- A typical measurement setup is

tion part (dependent on the modula- shown in Fig.3. The measurements m i , , . . tion frequency) and the second term is were carried out in an office building. there has been great interest in ^ • , i • v • • , j , 1 - - ^ the noise part, which is independent speech privacy tor many years. Cavan- r i L , , ,. £ „ „ 1 , 1 , 1 ., , , A • or the modulation frequency. When augh et ai have described how Articu- ,, - 1 4 . • ?• /a /xn • 1 ,. T 1 /A™ • the signal-to-noise ratio (b/.N) is lation Index (AI), average transmis- , ,, , c J O +U J J. , 1 1 1 1 • greater than 15 dB, the second term sion loss and background noise are 1 . . ,, , , ,, „• , , , 1 , , . 1 TT equals 1, meaning that only the first used to calculate privacy . However, , . , C/XT 1 +u a , m . . «, /nr-™- • term is measured. b/JN greater than or bound transmission Class (bIC) is , , - - JTJ . , , . , , ^.„ ,, , . , , , . equal to 15 dB is obtained by using still the most widely used privacy ,, , , , , ,, , ,, , ., elevated levels. measurement method even though it

has been shown that STC is not a very mi J , J • , ,. , r . Ihe second term, corresponding to good predictor for privacy. c , , , , , , « J reference levels, is calculated from

rm £ J. ii_ x • & u measurements of background noise LK The factors that influence speech . ,. . L , . ,' , T w

. , ,,. .,.,., 1 ,, c 1 and the elevated signal level Lc+N us-lntelligibihty and therefore speech . , _ , - , , . , 1 . , , . ,. n ing the following formula. privacy are signal-to-noise ratio, re- ° ° verberation, and reflection. The two methods used now, AI and STC, are _ IQLS+NAO _ IOLN/IO based on signal-to-noise ratio whereas ^' ~ ^QLN/10 the RASTI method takes all factors into account2. RASTI also measures A t y p i c a l R A g T I m e a s u r e m e n t pro-speech intelligibility according to IEC c e d u r e . g h o w n . Y % R A g T I fa

DLU.^OO . calculated from the nine modulation reduction factors as follows:

Theory x. = 10 log [m./(1 _ mj)]

The RASTI method is based upon a , v . ,, , , „ ,, T i ^ j i ^ - m i ? where X{ is the apparent signal-to-measure of the Modulation transfer . , . ,. , ,, ■1 ,• /x m-r x i • • noise ratio corresponding to the mea-Function (MTF) and gives a numeric , , , ,. , ,. -' ,, , , n j - /T^. .x .,, sured modulation reduction factor m;. result between 0 and 1 (rig. 1), with x

the range of interest for speech priva- truncated at X -i . r. i /-« r« m i 1 « 1 1 l i e J\.{ V a l u c b d i e ; L lUl lCdLcU. (XL yv; — cy between 0 and 0,3. The test signal , 1 e J T )

l , ,, , J , . , , , , . , . « i ± 15 dB such that used is modulated pink noise, .bor low RASTI values the resultant modula-

Fig. 1. The RASTI Scale for Speech Intelligibility Fig. 2. RASTI Flowchart

BO 0262-11

The partition between two workspaces I I „ I "REF + 30 dB I Conclusion was 1/2 inch gypsum board on a steel R E F corrected" frame with 3 inches of mineral wool in 0_. ,c™ n v 777^ 7T^ Ti u u u u i *. J r ;. , STI (500 Hz) 0,53 0,50 It has been shown how elevated lev-f h p PPT1TPT* __

STI (2 kHz) 0,12 0,02 e*s c a n ^ e u s e (* ^or P rivacY measure­rs . , . ,. . ments at the low end of the RASTI To make measurements in this situ- RASTI 0.30 0,23 scale which otherwise would have been

ation, it was necessary to raise the sig- T01793GB0 • -ui A i w J ± i i i i on m o- i • • , impossible. A calculation procedure to nal level by 30 dB. Since this is not Table. i. Measurement results -REF' is the r e l a t e t h e r e s u l t s t o «Ref> l e v e l i s possible with the Transmitter lype result obtained using the Type -i ■> n ,., ,. . ( ,> n ,, , n n r , , , , o o c ,. , , . ., n , C T 7 Y vided. Calibration information for the 4225, an external speaker was used 4225 adjusted to the RASTI refer- , o n , . (Sound Source Type 4224). This ence level.'REF + 30 dB, correct- B&K Sound Source Type 4224 is sup-source was calibrated by placing a mi- ed"ls the resf °?f « e d fr°m ^ p U e C t

, . ,., ,. , measurement with the elevated crophone on the calibration attach- level and then calculated to corre-ment supplied with the Type 42244. In spond t0 the standard level xCavanaugh, W. J., et al "Speech Pri-this case, the levels measured were vacy in Buildings", Journal of the

situation will overestimate the RASTI Acoustical Society of America, vol. 34 500 Hz 80,4 dB value, while the method described in no. 4, 1962.

2000 Hz 73,0 dB this Application Brief gives a more ac- 2Type 3361 RASTI Product Data Level Difference -7,4 dB curate result. The largest difference sheet, Briiel & Kjaer

between the two measurement proce- 3IEC Std.268-16, The Objective Rat-The measurement results are shown dures is in the 2000 Hz octave band; ing of Speech Intelligibility in Audito-

in Table 1. the MTF is shown in Fig. 4. Note that ria by the RASTI Method the m-values of the background noise 4Type 4224 Instruction Manual,

These results show that a measure- are higher than the m-values from the Briiel & Kjaer ment using a reference level in this measurement with "REF" level.

Fig. 3. Measurement Setup Fig. 4. The measured Modulation Transfer Function