Finding the right .sofa HRTF file?

[rbaudoux]

Hello,

I am currently setting up Spat to move sounds using the KNN algorithm. I want to monitor the output on headphones using spat5.virtualspeakers~ (binaural output). However, beyond the sampling frequency, I have no idea how to choose a .sofa HRTF. Indeed just the IRCAM database linked from Spat itself offers several hundreths of files even when restricted to 44.1kHz!

How can I find my way among these files?

[noone_511]

Hello, I’m working with Spat in VR and had the same issue, going through all the HRTFs and finding your best one seems the only solution so far. Or you can use this https://www.earfish.eu/ to get your own HRTF. I also found this GitHub - jaakkopasanen/Impulcifer: Measurement and processing of binaural impulse responses for personalized surround virtualization on headphones. but I haven’t tried it yet.

[rbaudoux]

Thanks @noone_511. Haven’t you tried Earfish neither?

[tcarpent]

Hi,

You’ll find some information here :

SPAT most genereral HRTF

Hi, Thanks for your comment. Categorizing suitable HRTFs is a hot topic of ongoing research. I assume you are aware of the BiLi project (http://www.bili-project.org). In this framework, we expect to assess quality of binaural listening and that could be of interest to you. Regarding Flux Spat, the following HRTFs has been recognized as « pleasant » by many users (including sound engineers and mixers) for they provide a rather « neutral » spectrum while preserving decent localization cues : …

Additionally:

• it is recommended to use “Compensated” HRTFs (a number of helpful post-processing operations have been applied).
• I recommend to use the “SimpleFreeFieldHRSOS” files. They contain HRTFs modeled as second-order-sections (SOS) which are CPU efficient and spectrally smooth. The file suffix SOS12, SOS24, SOS36 refers to the “quality” of the SOS modeling. SOS24 is typically a good choice.
• The dataset 1040, which is frequently well-appreciated, is now built into all binaural objects. Use /load IRC_1040 to load it.
• Make sure the sampling rate of the chosen file matches that of Max. (otherwise you’ll notice a warning in the Max console)

Best,
T.

[noone_511]

Not yet, but they are from the University of Antwerp so I guess the process is sound.