BinauralSDM

Matlab code to generate binaural RIRs for arbitrary head orientations using the Spatial Decomposition Method [1], [2] using the BinauralSDM approach with RTMod+AP equalization from [3].

Installation and dependencies

The following dependencies are necessary for the repository to run successfully. Please make sure that they are included in your Matlab search path before executing any demo:

• SDM Toolbox for Matlab (must be downloaded manually) - by Sakari Tervo and Jukka Patynen.

• SOFA API for Matlab (must be downloaded manually) - by the SOFA conventions team.

• getLebedevSphere.m (included) - by Robert Parrish.

• parfor_progressbar.m (included) - by Daniel Terry.

• denoise_RIR.m (included) - by Densil Cabrera and Daniel Ricardo Jimenez Pinilla.

• Matlab Signal Processing Toolbox - by Mathworks.

• Matlab Statistics and Machine Learning Toolbox - by Mathworks.

• Matlab Curve Fitting Toolbox (for denoise_RIR.m) - by Mathworks.

• Matlab Parallel Computing Toolbox (optional, parallelization will be skipped if unavailable) - by Mathworks.

Demos

Set your Matlab working directory to ./Src/Examples/ to execute the examples:

• Demo_BinauralSDM_QuantizedDOA_andRTModAP.m Generates BRIRs for a multitude of head orientations using the RTMod+AP equalization approach. To ensure that the example runs, it performs the analysis and synthesis of an example RIR measured with the FRL Array (10 cm diameter) and using TDOA analysis. The HRIR dataset (Neumann KU100) is downloaded on the fly from the TH Koeln Audio Group server. This can be easily swapped by any arbitrary HRIR dataset in SOFA format. This example generated direction dependent early reflections and direction independent late reverberation, after an arbitrary (configurable) mixing time. The example also includes spatial quantization of the DOA information. The file can be modified in a straightforward manner to accommodate other analysis and rendering parameters and input data.

More examples will be added in the future, featuring modified decay slopes for reverberation time manipulations, algorithmic late reverberation, and spatial manipulations.

Microphone arrays

Files for 3D printing of a microphone array holder (FRL Array) are included in ./Data/ArrayDesigns/. These are hexahedral arrays (6 DPA 4060) with a center microphone (Earthworks M30/M50) and a diameter of 5 or 10cm.

Other array geometries can be accommodated by modifying the file ./Src/create_MicGeometry.m. The current code also accommodates Tetramic and Eigenmike arrays (with TDoA estimation) but we recommend to not use them (see [3] for details and justification).

Citing BinauralSDM

If you use this code in your research, please cite the following paper:

@article{amengual20BSDM,
  title     =     {Optimizations of the Spatial Decomposition Method for Binaural Reproduction,
  author    =     {Sebastia V. Amengual Gari and Johannes Arend and Paul Calamia and Philip Robinson},
  journal   =     {Journal of the Audio Engineering Society},
  volume    =     {68},
  number    = 	  {12},
  pages     =     {959 -- 976},
  doi       =     {https://doi.org/10.17743/jaes.2020.0063}
  month     =     {12}
  year      =     {2020}
}

References

[1] S. Tervo, J. Patynen, A. Kuusinen, and T. Lokki, “Spatial Decomposition Method for Room Impulse Responses,” J. Audio Eng. Soc., vol. 61, no. 1/2, pp. 17–28 (2013 Jan.).

[2] S. Tervo, J. Patynen, N. Kaplanis, M. Lydolf, S. Bech, and T. Lokki, “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 Nov.), https://doi.org/10.17743/jaes.2015.0080.

[3] S. V. Amengual Gari, J. Arend, P. Calamia, P. Robinson, “Optimizations of the Spatial Decomposition Method for Binaural Reproduction,” J. Audio Eng. Soc., vol. 68, no. 12, pp. 959-976 (2020 Dec.), https://doi.org/10.17743/jaes.2020.0063.

Changelog

2022-08-30 - v0.5

• Add Save_BRIR_sofa.m to export rendered BRIRs as SOFA-file (activated by default over exporting individual WAV-files with according changes for new flags in create_BRIR_data.m)
• Rename SaveBRIR.m into Save_BRIR_wav.m
• Add Initialize_SOFA.m to extract functionality from Read_HRTF.m to be reusable
• Remove and add Align_DOA.m and Split_BRIR.m to fix git capitalization issues

2022-04-07

This update introduces changes to the names and parametrisation of internal functions which may break compatability to code using former versions of this toolbox. When applying the method it is therefore strongly advised to start by applying individually required modifications to the provided Demo. The script has been improved in terms of documentation, variable naming, logging verbosity, plot generation and data export.

• Update README.md with improved formatting, links to publications and changelog
• Update all functions to be more verbose by cleaning up and adding logging messages
• Update all function headers to follow consistent parameter documentation (e.g. create_BRIR_data.m and create_SRIR_data.m)
• Update all functions to follow consistent code formatting
• Rename function names to follow a more consistent convention:
  • align_DOA.m -> Align_DOA.m
  • ModifyReverbSlope.m -> Modify_Reverb_Slope.m
  • removeInitialDelay.m -> Remove_BRIR_Delay.m
  • split_BRIR.m -> Split_BRIR.m
• Update Align_DOA.m to use the beforehand estimated DOA of direct sound instead of averaging again (the former implementation as well as other methods are documented in the function)
• Add Apply_Allpass.m to extract functionality from Demo_BinauralSDM_QuantizedDOA_andRTModAP.m for modifying late reverberation
• Update Demo_BinauralSDM_QuantizedDOA_andRTModAP.m to skip downloading HRTF if already present
• Update Demo_BinauralSDM_QuantizedDOA_andRTModAP.m to use adjustable sampling frequency for allpass filters
• Update Demo_BinauralSDM_QuantizedDOA_andRTModAP.m to skip redundant exports of late reverberation
• Update Demo_BinauralSDM_QuantizedDOA_andRTModAP.m to generate verbose plots for all analysis, processing and export steps
• Update Demo_BinauralSDM_QuantizedDOA_andRTModAP.m to provide user-defined DOA rotation before DOA quantization
• Update Modify_Reverb_Slope.m to provide experimental RTmod regularization (deactivated by default)
• Add Plot_BRIR.m to generate and export a plot of a BRIR in (0, 0) deg direction
• Add Plot_DOA.m to generate and export a plot of SDM DOAs after SRIR analysis
• Add Plot_Spec.m to generate and export a plot of SDM spectra after SRIR analysis
• Update PreProcess_P_RIR.m to skip denoising if the Matlab Curve Fitting Toolbox is unavailable
• Update PreProcess_Synthesize_SDM_Binaural.m and Synthesize_SDM_Binaural.m to stack left and right HRIRs in single variable
• Update Read_HRTF.m to give instructions if the SOFA API is unavailable
• Add Rotate_DOA.m extracting functionality used in align_DOA.m and Synthesize_SDM_Binaural.m
• Update SaveBRIR.m to make export of combined and separate direct sound and early reflection optional
• Update Synthesize_SDM_Binaural.m to remove default specification of target BRIR length
• Update create_BRIR_data.m to provide additional options for reverberation equalisation process
• Add roty.m and rotz.m to eliminate dependency for Phased Array System Toolbox

2021-11-17 - v0.1

This update allows the user to specify two new fields in BRIR_data:
BRIR_data.BandsPerOctave: Specifies the frequency resolution for the reverb equalization. The values can be 1 or 3. For smaller rooms, 1 is recommended. By default it is 3.
BRIR_data.EqTxx: To perform reverb equalization, the RT60 must be estimated. However, there is generally not enough SNR to obtain a true T60 estimation. This parameter allows the user to specify the desired Txx for the RT60 estimation. Generally, a value of 30 is recommended. However, this can blow up the RT estimation in cases where SNR is low. For small rooms and very dry spaces we recommend using a value of 20. By default it is 30.

• Update Demo_BinauralSDM_QuantizedDOA_andRTModAP.m, GetReverbTime.m, ModifyReverbSlope.m PreProcess_Synthesize_SDM_Binaural.m, getLundebyFOB.m and getLundebyRT30.m with mew equalization options to make reverberation equalisation process more robust
• Update create_BRIR_data.m and PreProcess_Synthesize_SDM_Binaural.m to provide additional options for reverberation equalisation process
• Update Read_HRTF.m to fix loading of FRL HRTFs
• Update create_FIR_eq.m to use linear instead of cubic interpolation
• Update create_MicGeometry.m to add custom FRL array
• Update SaveBRIR.m, SaveRenderingStructs.m and create_SRIR_data.m to add custom path
• Update read_RIR.m to perform if sampling frequencies are mismatched

Contributing

See the CONTRIBUTING file for how to help out.

License

BinauralSDM is CC-BY-4.0 licensed, as found in the LICENSE file.

Contact

Sebastia V. Amengual (samengual@fb.com)

Philip Robinson (philrob22@fb.com)