The model requires a binaural input (left and right ear signals) to operate. This is in contrast to other measurement techniques that either analyse loudspeaker signals directly or use an arrangement of microphones to capture the acoustical signal under test. By starting with a signal that is similar to the type of signal that a human receives, this gives the model a better chance of giving a similar result to human perception.
Creation of a binaural signal
There are a large number of ways in which the binaural input can be created. A number of these are outlined below.
Recording of a binaural signal
The simplest method of obtaining a binaural signal is to make a binaural recording of the sound field within a concert hall or reproduction system using microphones placed in the ears of a human head. This is most commonly done using a head and torso simulator (HATS) or 'dummy head'. Such a device allows for the microphones to be placed within the ear canal, and it removes potential problems of accidental movement or noise from using a human head.
Dummy heads are expensive to buy and difficult to transport, so more practical and cost-effective alternatives were evaluated. We found that reasonably accurate results at all apart from high frequencies could be obtained by using a pair of small omnidirectional microphones positioned at the entrances to the ears of a human listener - though care had to be taken to reduce movement and noise. Reasonably accurate results could also be obtained at low frequencies by using a pair of spaced cardioid microphones.
Impulse response measurement and convolution
Even if the source signal and the reproduction system or acoustical environment are not available at the same time, it is possible to make measurements of the acoustical or electroacoustical system and then simulate the source signal being passed through the system by the use of convolution.
For instance, if the aim of the measurement is to test the spatial attributes that will be perceived when certain loudspeaker signals are reproduced within a room, then it is possible to derive the signal that will be heard at the listening position without needing to capture the signal directly. Firstly, the impulse response from each of the loudspeakers to each ear of a HATS can be measured. Then the loudspeaker signals can be convolved with the measured impulse responses to create a binaural signal that is suitable for measurement. In this way, measurements can be made more quickly, simply and without requiring access to the reproduction room. This technique can also be used to make measurements which mimic the reproduction of the loudspeaker signals over a number of different loudspeaker arrangements and types, and in different rooms, limited only by the availability of the measured impulse responses.
This technique can also be used to analyse the properties of auditorium acoustics. In this case, impulse responses are measured from one or more source positions within the acoustical environment to one or more receiver positions. Then, any number of source signals can be convolved with the measured impulse responses in order to create a binaural signal for analysis.
The importance of level calibration
As it has been found that the sound pressure level (SPL) of a sound has an effect on the perceived source width [Morimoto and Iida 1995
, Mason and Brookes ???
], the binaural signal that is input to the model must be calibrated to a known reference SPL. This is most simple to undertake if the binaural signal is captured using a HATS that has a calibrated input level. This is more difficult when creating suitable input signals by measurement of impulse responses and convolution, however it can be achieved if care is taken at all stages.