HRTF and personalization

HRTF? What is it?

When a sound wave travels from a given location to our eardrums, it interacts with several parts of our body such as the head and the ears. All these reflections, diffractions and absorptions (depending mainly on the shape of our head and the structure of our outer ears) change the nature of the incoming sound according to its direction of origin. These interactions between sound and morphology can very well be represented as direction-dependent filters. These sound wave filters depend on the source’s position in space. Our auditory system has learned over time to associate each specific filter to its origin. This is how we recognize where a sound comes from. The representation of these directional changes as filters is called “Head Related Transfer Function” or HRTF.

If you want to simulate a sound coming from a given direction, you must filter it with the HRTF of this direction, play it with headphones at the entrance of the ear canal and the auditory system will position the sound as coming from this spot in space since it recognizes the directional signature. This is the basic principle behind all the Virtual Reality Audio engines.





HRTF are personal!

HRTF filters are based on our morphology and we know that we are all different from each other with respect to our features. Therefore, HRTFs differ from one individual to another! If you use somebody else’s set of HRTF, you end up with inaccurate or fuzzy sound localization.

Benefits of using personalized HRTF

Existing VR audio solutions do a somewhat good job at virtualizing 3D Audio with non-personalized HRTF. However, anybody who has experienced a VR Audio content with his own set of HRTF knows that the perception improvement compared to non-personalized HRTF is massive! With your own HRTF, everything sounds clear, sharp, outside the head and incredibly real. You can really experience auditory presence in an audio scene. The best analogy we could think of is a visual one: listening to VR Audio with personalized HRTF is like watching video in UltraHD and listening with non-personalized HRTF is like watching a low definition blurry video (see images on the left).

 Personal HRTF by 3D Sound Labs

At 3D Sound Labs, we believe that HRTF personalization will be the next revolution regarding 3D audio experience.
The state of the art of HRTF capture is the acoustic method. The procedure is as follow: the user seats without moving in an anechoic room, with little microphones in the ear canals while a set of loudspeakers produce “frequency swipes”. The concept is that the microphones will capture the HRTF in as many directions as possible. There are of course many flavors of this process. At 3D Sound Labs we have a sound stage with multiple loudspeakers.

The “acoustic” method to capture HRTF is far from optimal for a consumer market: it is complex, expensive, hardly transportable and provides a rather limited number of sampling direction. Besides, the procedure is long and very uncomfortable for the end user.
That is why we introduce another solution: for audio professionals, we have developed a proprietary HRTF acquisition process based on high definition pictures where the scanning stage takes only 15 minutes. It provides very accurate high resolution HRTF, is transportable and is an excellent alternative to the state of the art way of capturing HRTF in an anechoic room.




Cracking the HRTF Personalization Challenge

Our vision is that you will soon personalize your VR headset just like you do today with your phone, when you configure for instance your fingerprint sensor. Tomorrow, you’ll take a few pictures of yourself to create your avatar in the VR world. Those pictures should be enough to personalize one’s HRTF, making this process totally seamless for the end user.

Our R&D team and its academic research partners have also been working on a consumer-friendly technology to provide this HRTF personalization to our 3D Audio Engine from pictures acquired from a regular smartphone.
3D Sound Labs is currently running a large-scale research program intending to generate personalised HRTF based on a series of photos.

We are currently building a very large database of 3D ear models that will enable our machine-learning-based process to extract personalized HRTF from a series of pictures. This work has been presented at the Audio Engineering Society convention in Los Angeles in September 2016.
You can read the related AES engineering brief below.

Read AES Paper


Efficient Process

The research program currently lead by 3D Sound Labs will ultimately bring an ultra-light, consumer-friendly and efficient HRTF personalisation process to the market.

This framework for individualizing HRTFs, requires very few inputs from the end user, which is a critical feature for industrializing the process. Moreover, this method is not time-consuming from the end user point of view as most of the computation is done offline, during the dataset creation and the models building.



How does it work?


The photographs are fitted to a statistical 3D ear model coupled to a statistical HRTF model. This coupling allows to instantly generate from a given parameterization of the ear model a corresponding set of HRTFs. The accuracy of the results is a direct consequence of the quality and the size of the underlying databases.

3D Sound Labs is currently building its own database of hundreds of ears and their related HRTF, and which is already set to be one of the most massive on the market. This massive database allows 3D Sound Labs to yield two statistical models, linked together for creating a coupled model of ears and HRTFs.
Finally, photographs of face and ears of the end user are processed in combination with the previous model to generate and adapt a brand new set of personalized HRTFs.

3D Sound Lab’s dramatic advantage here is that it is working with both calculated and measured data. Calculated data allow to work on a large scale with limited ressources and time consumption (more flexibility, repeatability) whereas measured data further validates the numerical data.


In all cases, 3D Sound Labs’ Personalized HRTF solution leads to far less sound localization errors than the standard KEMAR HRTF.