Disentangling Parametric Spatial Room Impulse Response Methods: Examples
This site is still under construction!
This page contains some examples of parametric SRIR processing. All examples are based on measurements of two variable
acoustics rooms. Room Impulse Responses were measured using different microphone arrays, including the GRAS VI-50 intensity probe, the open array used in [5],
the Eigenmike em32, and the tetrahedral cardioid array Rode NT-SF1. This allows for comparison between some of the methods that are summarized in the table below.
Note that these renderings can only be regarded as snapshots; they are examples that we have produced while reviewing different methods.
It is certainly possible that better parameters or other measurements give better results. Also, as methods inevitably use different microphone arrays
and rendering methods, one can not make general statements by directly comparing these renderings in all regards.
Different Approaches
Here we provide renderings of the same room, created with different parametric SRIR workflows. HO-SIRR, SDM and ASDM can be used for loudspeaker rendering.
To take the full reproduction chain into account, the responses were rendered to a 37 loudspeaker setup placed in an anechoic room at the Aalto Acoustics lab.
In order to present results online, the loudspeaker responses were then convolved with a set of BRIRs, measured at the center of the array.
- SDM uses the SDM toolbox, following the demoCostum script. With the settings suggested there, some artefacts can occur due to the equalization, see next section.
- Measurement: Open array of omnidirectional microphones (incl. central microphone)
- Analysis: Broadband TDoA estimation, equalization
- Rendering: NLS to 37 loudspeaker, binauralization using KEMAR HATS BRIR measured in the loudspeaker array
- HO-SIRR uses the provided code
- Measurement: Eigenmike m32, 4th order encoding using analytical filters with Tikhonov regularization
- Analysis: Narrowband, sector-based intensity vector, diffuseness estimation, broadband rendering of direct sound
- Rendering: VBAP to 37 loudspeakers, dedicated directional diffuse rendering, binauralization using KEMAR HATS BRIR measured in the loudspeaker array
- ASDM was implemented using the code provided in [3].
- Measurement: Soundfield microphone
- Analysis: Broadband PIV estimation, filterbank equalization
- Rendering: Ambisonics encoding to 4th order, mode-matching decoder to 37 loudspeakers, binauralization using KEMAR HATS BRIR measured in the loudspeaker array
- [5] desribes a binaural method based on SDM. For this, the responses were not rendered to the loudspeaker array, but HRIR measurements of the KEMAR HATS were used directly.
- Measurement: Open array of omnidirectional microphones (with central microphone)
- Analysis: Broadband TDoA estimation
- Rendering: Direct binaural rendering using KEMAR HATS HRIR, slope correction (RTMod) in fractional octave bands and Allpass (AP) roughness compensation
- The Binaural condition is a reference measurement using the KEMAR HATS. The same diffuse-field EQ as for the virtualized loudspeaker playback was used.
Scroll sideways to find more samples.
Specific Aspects
Equalization for Broadband Methods
These examples were created using TDoA-based broadband directional estimation,
nearest loudspeaker rendering to 37 speakers and subsequent binauralization, such as typically done in SDM [1].
Also the measurements used here were done in a variable acoustics room, however taken from a different dataset than the examples above.
In such a workflow, coloration compensation should be applied. These examples demonstrate different coloration compensation methods.
Using the settings suggested in the demo script found in the SDM toolbox can sometimes lead to artefacts, which might explain some of the problems other researchers have observed.
Different Equalization methods were tested, such as filterbank-based equalization, similar to [2]. It was applied to loudspeaker responses instead of Ambisonics channels.
The spectral differences are perceived also in case of the continuous cello signal. Roughness is only heard in case of the transient kickdrum sample (which was also used in [4]).
Scroll sideways to find more samples.
Reverberant Room
Dry Room
HO-SIRR with varying input order [4]
The following examples were rendered using the complete HO-SIRR [4] workflow. The same variable acoustics room as before was used, but this time the Eigenmike em32 was used instead of the GRAS intensity probe. With a measurement from a real array, some spectral differences are immanent between varying input orders.
Reverberant Room
Scroll sideways to find more samples.
Dry Room
Scroll sideways to find more samples.
Examples from other studies
The signals used for the listening test described in [3] can be found here here
More examples of HO-SIRR can be found along with the sourcecodeReferences
[1] S. Tervo, J. P. Tynen, A. Kuusinen, T. Lokki, "Spatial Decomposition Method for Room Impulse Responses," J. Audio Eng. Soc., vol. 61, no. 1, pp. 17-28 (2013 Mar.).[2] S. Tervo, J. Pätynen, N. Kaplanis, "Spatial Analysis and Synthesis of Car Audio System and Car Cabin Acoustics with a Compact Microphone Array,"
J.Audio Eng. Soc, vol. 63, no. 11, pp. 914-925 (2015 Feb.), https://doi.org/10.17743/jaes.2015.0080
[3] M. Zaunschirm, M. Frank, F. Zotter, "Binaural Rendering with Measured Room Responses: First-Order Ambisonic Microphone vs. Dummy Head,”
Applied Sciences, vol. 10, no. 5, p. 1631 (2020 Feb.), https://doi.org/10.3390/app10051631
[4] L. McCormack, V. Pulkki, A. Politis, O. Scheuregger, M. Marschall, "Higher-order Spatial Impulse Response Rendering: Investigating the perceived effects of spherical order, dedicated diffuse rendering, and frequency resolution," J. Audio Eng. Soc, vol. 68, no. 5, pp. 368-354 (2020 May), https://doi.org/10.17743/jaes.2020.0026.
[5] S. V. A. Gari, P. T. Calamia, and P. W. Robinson, "Optimizations of the Spatial Decomposition Method for Binaural Reproduction," J. Audio Eng. Soc., vol. 68, no. 12, p. 18, 2020.