Impulse responses (IRs) are recorded by either firing a loud impulse such as a starter pistol and recording the resulting acoustic stimulation, or more normally these days by using a sweep sine tone method. IRs in this project were recorded from three different locations.
Equipment used:
- Genelec 1031 loudspeaker
- SPS200 microphone (in later installations replaced by the MhAcoustics EM32)
- Sounddevices 788T recorder (for mic preamps)
- MOTU828MkII audio interface
- MacBook Pro
- MaxMSP8 and the HISSTools Impulse Response Toolbox MLS and Sinetone sweep method (later replaced by the Spat5 sweep measurement kit).
(In later project modules the SPS200 was replace by the EM32 microphone recording onto a laptop using the Spat5 sweep measurements kit).
3D IRs are musically interesting because they can be used to make a dry sound appear as if it is sounding in the space from which the IR was made, and also because by increasing the volume of what is in fact a reverberation emulation, we can hear acoustic features that were less obvious in natural listening. Temporally they are also interesting because they capture macro space and micro time (rather than macro time as in a normal recording). This means that by extending the time of the recording (by either audio processing or numerical analysis) the boundary between space and time begins to blur. More about this will be discussed in the section on creative work.
Although impulse responses reveal acoustic fingerprints, what is captured is only true for the specific locations of loudspeaker and microphone used in the recording. We should also remember that the loudspeaker creating the sound-source is directional.
To use 3D IRs musically it is useful to overcome the problem of the unique location connecting the microphone and impulse. Modelling the complete acoustics features of a space from a collection of IRs is far from trivial and is an incredibly active field of research in science and engineering. At the time of writing there is not yet a practical solution of sufficient quality to use artistically. Instead I developed my own approach, which was perceptually tested by science collaborator Franz Zotter and myself against a method that applies the Ambisonic Spatial Decomposition Method (ASDM). We found my results to be perceptually more satisfying, although likely mathematically less precise. (By the time the Reconfiguring the Landscape project has ended, there may be a better solution available).
Example 2: Ambisonics recording is convolved with the three ambisonics IRs taken from different locations and directions of face, rendered as binaural for this example. Here I also apply the correct matrix convolution. Although the combination of three IRs was not correct from a technical standpoint, it begins to reflect more accurately the real scene.
Example 3: Here I decode the ambisonics IR and the ambisonics recording to a virtual loudspeaker array, convolve each of the decoded channels one to one, and then re-encode the results in HOA. The binaural example shows that this is by far the best option. There are two features to note:
- When the source directions are placed below the horizontal there is a spatial distortion. This is because I chose to decode both IR and source sound to a virtual hemisphere (to reduce the number of convolution channels for this test). A full sphere will be more appropriate if the source that is being convolved is also spherical.
- When the source direction is frontal we hear a gain increase. The bias reflects the original spatial relationship between the microphone and the loudspeaker in the IR recording. This gain increase can then be compensated for by spatially attenuating in the ambisonics domain using one the Ambix Directional Loudness vst plugin.