ILD(interaural level difference)

Spatial hearing tutorial

Introduction

Some important aspects of spatial hearing are evident from the sound received at the two ears, together with the interaural differences.

In this tutorial, a physical model of a human (known as a head and torso simulator) is used to record sound in order to make impulse responses from various distances and directions in a medium-sized room. Impulse responses made in a room at the two ears are known as ‘binaural room impulse responses’ (often abbreviated to BRIRs).

In addition to the measurements demonstrated in the tutorial, the tutorial also provides a set of BRIRs made in various places, ranging from a very small room the Sydney Opera House concert hall.

With binaural impulse responses, listening is very useful in understanding the relationship between perception and measurements.  If you listen to the measurements using headphones, then your ears pick up what was received at the two microphones of the head and torso simulator.

Measurements in class in 2021

Measurements in class were made with a small loudspeaker and a head and torso simulator.

I’ll add a video of this part of the class here when it is ready – in the mean time the lower quality zoom video is available in the Week 11 module.

These measurements are to be used in assignment 2. Access the AARAE results from the class here  download. (this file is sent to the customer support email)

The measurements were made at various distances and azimuth angles. The suggested positions were the following, but some other ones were done too. Watch the video to see what we did.

1 m, 0 deg

1 m, 45 deg

1 m, 90 deg

1 m, 135 deg

1 m, 180 deg

0.25 m, 0 deg

0.25 m, 90 deg

2 m, 0 deg

4 m, 0 deg

Converting raw measurements into impulse responses

The measurements were made using a sinusoidal sweep (exponential), which is the most common test signal for making impulse responses in audio and acoustics. Measuring an impulse response using this technique is advantageous because (i) the long duration sweep creates a lot more energy than a short impulse ‘click’, thereby dramatically increasing signal-to-noise ratio, (ii) the exponential sweep separates harmonic distortion from the linear response, and (iii) the exponential sweep is not badly affected by system time-variance during the measurement (although that is not an issue in these measurements).

Each recording includes the sweep’s inverse filter in a secondary field (called audio2); an impulse response is created by convolving the sweep recording with the inverse filter. You can do this by using the ‘*’ (star or asterisk) button in AARAE. Part of this process is a dialog box that allows you to truncate the audio (we only want the linear impulse response, which is a small part of the recording just after its mid-point). For the purposes of this exercise, we could either choose ‘causal part …..’ to preserve the absolute timing of the impulse response (including system latency) or one of the peak-finding methods.

Listening to the impulse responses

When you have a binaural impulse response, you can listen to it directly or with further processing. AARAE gives you a few options for listening, including forwards, reverse, random phase, flat magnitude, and convolved with ‘I’m Speaking from Over Here’. These playback methods are all normalized, which means that you won’t hear the effect of reduced direct sound level with greater distance. Notice – if you play the IR backwards the reverberation is much more noticeable (the auditory system de-reverberates the sound of rooms, to some extent).

Also, you can edit the impulse responses, e.g. to only include the direct sound. This is done using AARAE’s edit button: change the end point so that it is just before the first reflection; also it might be a good idea to change the start point so that it is just before the direct sound (if ‘causal part’ was used in sweep processing). In order to find the right start and end points, create a chart of the IR and zoom in. Then open the edit GUI and apply the new start and end points, edit the Name, and ‘Write to new’.

BTW – if you are interested in more advanced detailed listening (of direct IR sound only – not including the reverberation), AARAE includes sonifyHRIR_CabreraMartens2011. This ‘zooms in’ and enhances the impulse response features, including the spectrum and time-envelope. This is described in: D. Cabrera & W.L. Martens, ‘Sonifying head-related transfer functions’, in  (Links to an external site.)Yoiti S, Douglas B, Kazuhiro I, editors. Principles and applications of spatial hearing. World Scientific; 2011 Mar 11, p473-483 (Links to an external site.). To summarize what it does:

• a ‘hiss’ is included that is equivalent to white noise convolved with the binaural impulse response (this is the same as AARAE’s ‘phi’ random phase button). This allows you to directly localize the sound source (to the extent possible).
• The binaural time-envelope (slowed by a factor of 100) is used to modulate an exaggerated binaural spectrum (transposed down by a factor of 10, and exaggerated using a magnitude exponent of 1.5). This allows you to analytically listen to the interaural time differences, level differences, and spectral cues.

Analysing binaural room impulse responses

AARAE’s analyser IACC_IRbands is probably the most useful analyser for the binaural impulse responses. It calculates interaural cross correlation functions and interaural level differences and associated parameters (as per conventions in large room acoustics). See ISO 3382-1 for a brief introduction to using IACC in auditoria. Here are some hints about the output of this analyser:

• Emeans ‘early’, which in auditorium acoustic refers to the first 80 ms of the IR (starting from the direct sound). You can adjust this to a different time in the analyser dialog box. You can even adjust it so that ‘early’ only captures the direct sound.
• Lmeans ‘late’, with refers to the sound after ‘early’.
• Ameans ‘all’, i.e. the whole impulse response including early and late parts.
• IACC(interaural cross correlation coefficient) is the maximum absolute value peak of the interaural cross correlation function (in each octave band). A value close to 1 indicates that the sound mainly comes from one direction; a value closer to 0 indicates that the soundfield is diffuse (but very low values are unlikely in the low frequency bands due to the size of the head vs the long wavelengths).
• tauis the time delay associated with the IACC value (the interaural time difference), which can sometimes provide an indication of lateralization.
• ILD(interaural level difference) is the sound energy level difference between the two ears, in dB.

Other questions that might be of interest with these impulse responses include:

• Does the direct sound follow the inverse-square law? This can be examined in a number of ways, the simplest being to read the peak of the level-time plot of the IR.
• What is the relative energy level of the impulse responses (& how does the reverberant field contribute)? There are lots of ways that this could be examined (e.g. the octave_band_level_barplot or LevelStats can be used in the familiar way; ReverberationIR2 can be used to discover reverberation characteristics; or the analyser ClarityandDefinition_varTime calculates clarity index with a wide range of evaluation ranges, including as low as 10 ms; etc).
• How do reflections in the room affect the impulse responses, and auditory spatial impression? In the 2020 measurements, measurements were made with the curtains over wall vs bare wall. In the 2021 measurements we tried out an acoustic retroreflector array.