Measure Speech Intelligibility with a Sound Level Meter

Sound and Vibration, Oct 2004 by Bjør, Ole-Herman

Assessment of speech intelligibility for an emergency signalling system or a public address system may be an important task for an engineer or consultant. Today, the analysis power of a modern sound level meter is sufficient for direct measurement of speech intelligibility using STI or CIS metrics.

Several factors determine how well speech is received when wo listen to a speaker. It depends on the performance of the speaker as well as on our psychoacoustical system. But, in most cases, intelligibility is characterized by the acoustics of a sound reinforcement system and associated listening space. For such applications we desire an objective assessment of the speech intelligibility, independent of the speaker and the listener. The speech transmission index, STI, has proven to be a valuable tool for such an objective assessment. From the first presentation in Acustica in 1971,1 the method has been refined and diversified for various applications. Recently the International Electrotechnical Commission, IEC, has launched the third revision of the International Standard specifying the method for calculating the STI index, IEC 60268-16.2

The STI methods can be used to compare speech transmission quality at various listening positions and under various conditions within the same listening space. In particular it is useful for assessing the effect of changes in acoustic properlies, including effects produced by the presence of an audience, changes in room surface properties or changes in a sound systern. The methods are also able to predict the absolute rating of speech transmission quality with respect to intelligibility when comparing different listening spaces under similar conditions or assessing a speech communication channel.

Speech Transmission Index - STI

The STI-method for assessing speech intelligibility was developed at TNO Human Factors in the Netherlands by Tammo Houtgast and Herman Steeneken. The basis for the STI index is that the intelligibility of speech is largely based on the slow amplitude modulation of a sound that acts as a carrier. In the STI-method, the carrier is stationary gaussian noise divided into seven octave bands ranging from 125 Hz to 8 kHz. The width of each band is one-half octave. Each of the bands is modulated with one of 14 modulation frequencies. The modulation frequencies are selected in one-third octave steps from 0.63 Hz to 12.5 Hz, which gives a total of 98 combinations.

In STI context, the term "intensity" is used to denote the square of the sound pressure Pa2. Intensity is the quantity being modulated. A small loudspeaker, roughly the size of a human mouth, driven by the modulated excitation signal, acts as a talker.

The sound at the listener position is received by a microphone. The level and degree of modulation in each octave band are used to determine the speech transmission index. Noise and reverberation in the room will reduce the observed degree of modulation. The method also considers the effect of the most common types of distortions such as harmonic distortion and intermodulation. However, some other forms of nonlinearity, like frequency shifts and frequency multiplications, are not treated effectively.

In order to take the effects of nonlinoarity into account, it is important that the basic signal being modulated is a random noise signal with a high crest-factor, a spectral distribution similar to the long-term speech spectrum and that the main modulation frequencies are selected one by one. The measurement of the full STI therefore has to be performed as a sequence of measurements. If each of tho 98 combinations is measured for 10 seconds, the total measurement time will he about a quarter of an hour. Such a long measurement time to make an STI measurement in only one position of a room limits the applicability of the full STI method. As a result, ways were developed to reduce the measurement time while giving almost equivalent results.

The STI method may be modified in different ways to reduce the time needed for the measurement. If the system to be measured is regarded as linear, then a number of solutions exist. The excitation signal may be modulated with all modulation frequencies simultaneously and the components may be separated after reception by the use of filters or by Fourier analysis. A more common method is to calculate the complex modulation transfer function m from the impulse response of the room from which the STI value can be determined.

The STI value is a weighted average of the different modulation indices. The last revision of IEC 60268-16 also considers masking effects and the absolute threshold of hearing.

RASTI and STIPA

In order to simplify direct measurements, the RASTI method (Rapid Analysis - or Room Acoustics - Speech Transmission Index) was developed at TNO in 1979. Different instruments were developed for the measurement according to this standard. A typical measurement time was 10 to 15 seconds. The RASTI method only considers two octave bands - 500 Hz and 2 kHz.