AMS Acoustics is a world leader in the field of speech intelligibility prediction.
The practice has a number of custom mathematical models to predict both STI and RASTI system performance. Models for both classical (Sabine) and non-classical spaces are available.
This service is particularly useful for verifying contractor's design prior to contract.
The science of speech intelligibility considers the transmission path between source(s) and receptor(s) and the subsequent degradation and corruption suffered by the speech signal en-route.
In terms of sound system design both the acoustics of the space and the system components and their disposition have a major effect.
To some extent the affect of acoustical and system factors are distinguishable. For example, the directional properties of the source (loudspeaker) from one viewpoint is an acoustical effect and the other is a system attribute. Acoustically, the directional properties affect the direct-to-reverberant ratio.
The direct component is that part of the loudspeakers output which arrives directly to the listener and the reverberant component is that part of the loudspeaker's output which serves little or no useful purpose.
Speech intelligibility increases with increasing direct-to-reverberant ratio.
This may be readily understood by considering speech communication in a large reverberant space, such as a cathedral.
At large distances between a talker and listener, intelligible communication is difficult. The reverberant sound would mask the speech syllables since the direct sound would be weak and the reverberant sound dominant. As the talker and listener move closer together, then the direct sound increases and speech communication improves. Clearly the direct-to-reverberant ratio is increasing and intelligibility improving. At very close distances, say less than 1 metre, the direct sound dominates and speech intelligibility becomes good. Intelligibility may be close to 100% even though the space has a long reverberation time.
If the distance between the talker and the listener further decreases such that the talker is able to whisper in the listeners ear, then there is virtually no audible reverberant component and the direct-to-reverberant ratio approaches infinity, and hence the intelligibility is no longer a factor of, and is not influenced by, the space.
The foregoing scenario is also applicable to public address/voice alarm systems. Different types of loudspeakers have different patterns, which are able to directly affect the direct-to-reverberant ratio.
THE STI/RASTI METHOD AND SCALE
The figure below shows the interrelationship and correlation between the various objective measurements and perceived speech intelligibility.
It can be seen that the objective intelligibility descriptors correlate with subjective word scores and hence to perceived intelligibility.
The STI/RASTI method uses amplitude modulated noise with a known modulation depth of the transmitted signal and analyses the return signal to assess the reduction in depth of modulation caused by noise and reverberation.
The figure below demonstrates this principle.
The STI/RASTI scale is between 0-1, the figure below puts the scale into perspective.
Of particular interest is the STI/RASTI value of 0.5 which is equivalent to 0.7 CIS, the requirement of BS EN 60849: Sound Systems for Emergency Purposes.
The figure below illustrates the importance of the direct component.
Clearly therefore, an understanding of both acoustics and sound system design is paramount.
The prediction of STI/RASTI requires the following data input:
- Geometry of the space
- Acoustics of the space
- Sound system design
- Noise levels
- Speech characteristics.
The figure below shows the input and output data to the AMS STI/RASTI computation engine.
It should be noted that for RASTI, only the 500Hz and 2kHz octave bands are used.
The actual prediction of STI/RASTI performances is carried out in the AMS STI/RASTI engine.
The input can take various forms, our preferred method is to input the direct field from an acoustical CAD model. This is the quickest and most effective method of combining the loudspeaker characteristics with the disposition of the loudspeaker throughout the space.
It should be emphasised that only the CAD model deduces the direct field. The figure below shows a typical coverage plot.
The reverberant field is calculated from standard formula with corrections applied depending upon the distribution of absorption and the aspect ratio of the space. Correction factors are also applied to take account of unidirectional flow, e.g. for non-classical spaces, tube platforms, openings, corridors, etc.
THE STI/RASTI ENGINE
AMS's proprietary STI/RASTI engine is a mathematical model which is visual basic driven using a spreadsheet input and output interface.
The engine itself derives the general solution to the modulation transfer function from the impulse response in the space.
In the first version (V1.0) the STI/RASTI engine provided results in the reverberant field only. This was superseded by (V.1.1) which used an empirical relationship to determine the effect of both direct and reverberant fields.
During 1994 V2.0 was developed which did not rely upon empirical correlation, and in 1995 V2.1 was produced which is Macro driven and provides a matrix output.
The AMS STI/RASTI engine (V2.0) has been validated on a large number of both acoustically classical and non-classical spaces. The deviation from the mean between predicted and measured values is within ±0.02.
There are no reported errors in the operation of the RASTI engine, there was one reported error in the supervising Macro, which has now been determined and repaired.
The AMS STI/RASTI prediction engine has London Underground Ltd. 'Approved Status'.
In 2001, the STI/RASTI engine was extended to include prediction of full and modified STI. The Common Intelligibility Scale (CIS) was also implemented to allow an indication of expected word score results.
The engine V3.0 uses the same formulae and corrections used in V2.0 and so approved status still stands.
Such is the speed of V3.0 that STI is now the preferred method of predicting and auditing speech intelligibility as this covers all frequencies rather than, as previously stated only the 500Hz and 2kHz octave bands.
The output format of V3.0 is as follows:
- Summary sheet including input data and percentage of floor space which does not meet some minimum requirements.
- Direct Level Coverage at any octave band 125Hz to 8kHz.
- STI/RASTI coverage presentation in tabular form.
- D/R ratio at any octave band 125Hz to 8kHz.
- S/N ratio at any octave band 125Hz to 8kHz.
- Total expected A-weighted level in tabular form.