The Open Binaural Renderer behind Eclipsa Audio
Democratising immersive audio
Text and illustrations: Katarzyna Sochaczewska
Block 1
Eclipsa Audio entered the stage of spatial audio technology with a clear objective: immersive audio for everyone. Its free, open-source tools blend into existing workflows, enabling delivery of spatial audio in an iamf format – currently supported by YouTube, the newest Samsung devices and soon also on Chrome browser and Android, with more platforms continually being added. Eclipsa is committed to every stage of immersive audio creation. In this article, we explore one particular dimension: Eclipsa's Open Binaural Renderer (OBR) – its architecture and rendering quality.
Eclipsa Audio is based on Immersive Audio Model and Formats (IAMF), an audio container developed by Google, Samsung, and other key contributors within the Alliance for Open Media (AOM), and released under AOM’s royalty-free license. This open-source approach represents more than just technological innovation – it’s a strategic move to eliminate the costly licensing fees that have long restricted immersive audio to high-budget productions. Eclipsa Audio represents a fundamental rethinking of how spatial audio should be structured and delivered. IAMF empowers creators and platforms to deliver high-quality, immersive audio with efficient encoding and flexible rendering, adaptable to various playback environments including headphones, multichannel setups, and VR systems. As a codec-agnostic format, IAMF is highly versatile for different use cases. It offers a rendering flexibility through “Mix Presentations” that translate to all possible playback systems, including binaural rendering powered by OBR, developed with a particular focus on perceptually outstanding quality.
Architecture
Eclipsa supports all existing formats – channel-based and Ambisonics, with object-based support coming soon – offering maximum flexibility in content creation. The architecture underlying how OBR combines and renders all audio elements is represented in the flowchart above. OBR supports up to 7 th-order ambisonics. It overcomes the upper frequency reproduction limit by implementing the Magnitude-Least-Squares (MagLS) solution for Direct Sound [1]. Head Related Impulse Responses (HRIRs) are used from SADIE II database [2].
Block 3
Central to the OBR’s spatial authenticity and externalisation is the virtual room implementation. In OBR-v1, it is simulated using MCRoomSim [3], a hybrid acoustic modeling system that combines the image source method with ray-tracing techniques. The virtual environment measures 5.2 × 5.8 × 4.3 meter with a reverberation time of 0.27 seconds in accordance with BS-1116 standards for critical listening environments. This space is rendered through an array of 86 virtual loudspeakers positioned at a 1.5 meter radius around a 7 th-order virtual ambisonics microphone receiver. This high-density speaker config- uration enables precise spatial resolution while the seventh-order Ambisonics capture provides the foundation for accurate soundfield reconstruction across the entire listening sphere.
Methodology of objective and Perceptual Evaluation of the OBR
The OBR represents more than a rendering engine – it embodies the philosophy of open, collaborative audio technology development. As the first comprehensive, production-ready implementation of IAMF binaural rendering, it establishes both technical standards and expectations for the wider ecosystem. Committed to trans- parency, it is undergoing both objective and perceptual evaluation as part of its ongoing development and continuous improvement.
The evaluation methodology includes objective and subjective components. Objective analysis covers key spatial audio metrics including frequency response, interaural level difference (ILD), interaural time difference (ITD), interaural cross-cor- relation (IACC), direct-to-reverberant ratio (DRR), reverberation time (RT60) and interaural coherence (IC), all for positions corresponding to a standard 7.1.4 speaker layout. Subjective evaluation is based on the overall listening experience (OLE) ratings, spatial and timbral fidelity attributes scores and qualitative feedback, all gathered using a diverse set of audio samples. The results are then compared against other existing binaural renderers. Each version of the OBR undergoes iterative testing, with insights from both objective and subjective evaluations informing continuous improvements in rendering performance.
Given the importance of preserving compatibility and ensuring accurate translation with existing audio format standards, the primary focus is on audio content produced in 7.1.4 format. Measurements were conducted by placing test signals at virtual loudspeaker positions corresponding to a standard 7.1.4 configuration. Impulse responses were measured using an exponential sine sweep (ESS) generated with a Python script. The ESS covered the frequency range of 10 Hz to 24 kHz, sampled at 48 kHz, with a duration of 6 seconds, a fade-in time of 0.25 and a fade-out time of 0.1 seconds, and a level of -12 dBFS, following an envelope of -6 dB/octave [4].
Block 5
The audio examples used for the evaluation spanned a broad range of material, including orchestral and acoustic music, speech and sound effects.
Initial renderer modifications
Initial analysis of the OBR renderer revealed a dip at 40 Hz, attributed to comb filtering in the virtual space, followed by some additional fluctuations below 100 Hz. To address this, a high-pass filter was applied above 100 Hz to the Binaural Room Impulse Responses (BRIRs), effectively reducing interference between the direct and reverberant sound. The result was a more consistent and flatter low-end frequency response, with improved overall clarity.
Block 7
Further analysis of the objective metrics of different binaural renderers is discussed in the context of the results of a subjective test. A MUSHRA-style interface was used to evaluate the Overall Listening Experience (OLE) across a range of different audio contexts and rendering methods, comple- mented by open-ended participant feedback. We present the OLE results for OBRv1 in figure 3. The results vary depending on the audio context, and are statistically significant for all examples, except for the Score example.
Block 8
The combined results from the objective metrics, MUSHRA testing and semantic coding analysis informed the redesign of the virtual room implementation and the optimisation of the direct-to-reverberant ratio. DRR provides key information about the perceived distance of a sound source from the listener. As the distance increases, the direct sound level decreases, resulting in a corresponding decrease in the DRR value. In binaural headphones reproduction, DRR plays a pivotal role in facilitating externalisation (perception of the sounds source appearing outside of the listener’s head). Optimal DRR values depend on the characteristics of the auditory scene and distance perception is further influenced by familiarity bias.
We also investigated the impact of Ambisonics order on the perceptual quality through two listening tests:
• Recognition test (ABX): Participants matched a sample (3rd or 7 th-order ambisonics) to a reference (7 th order).
• Preference test (AB): Participants selected the preferred version (3rd vs 7 th order), based on Overall Listening Experience.
No statistically significant differences in Test-Retest reliability were observed for either the recognition or preference surveys. This suggests that perceptual differences between Ambisonics orders are primarily perceived by a small population of highly skilled or experienced listeners and are more prominent in the recognition test rather than in the Preference test. Interestingly, we observed an overall preference for the lower Ambisonics order [5].
Block 9
Higher order filters reproduce the notches and peaks of the Head Related Transfer Functions (HRTFs) with higher precision. The differences between the orders are particu- larly noticeable in the higher frequency range, for example peaks around 1.7 kHz, and 7 kHz in case of the 7th-order ambisonics. For some listeners, whose individual HRTFs significantly differ from the HRTF set used in rendering, this can compromise clarity, narrow perceived width, or cause sound sources with strong harmonic content (such as percussive elements) to tilt from the center towards the left. Using lower order may smooth out these spectral features, potentially providing a more consistent and preferable experience for a broader audience.
Future of Eclipsa
The theoretical capabilities of OBR and IAMF are being validated through growing industry adoption. Since January 2025, creators are able to upload videos with Eclipsa Audio tracks directly to YouTube. Samsung is leading the way as the first industry adopter, integrating Eclipsa Audio across its entire 2025 TV lineup – from the Crystal UHD series to the premium flagship Neo QLED 8K models. The next version of OBR is scheduled for release in Autumn 2025.
References
[1] Rudzki, T., Kearney G., Skoglund, J. (2025): On the Design of the Binaural Rendering Library for Eclipsa Audio Immersive Audio Container, 158th Convention of the Audio Engineering Society
[2] MCRoomSim: https://github.com/Andronicus1000/
[3]: https://github.com/google/obr
[4] Farina, A. (2000): Simultaneous measurement of impulse response and distortion with a swept-sine technique. Preprints-Audio Engineering Society
[5] Upton, G. J. (1992): Fisher’s exact test. Journal of the Royal Statistical Society: Series A (Statistics in Society), 155(3), 395–402.
Block 12
Katia Sochaczewska, PhD, is an immersive music producer and researcher specialising in spatial audio technologies. She currently leads the perceptual evaluation and optimisation of Eclipsa Audio’s Open Binaural Renderer at University of York. She completed her PhD from AGH University of Science and Technology in Kraków, Poland, focusing on perception in ambisonics technology. Her immersive compositional and production skills are backed up by deep understanding of auditory perception. Beyond production and research, Katia is dedicated to sharing her knowledge and creating accessible resources such as the ECHO Project – to help composers, music producers and engineers explore immersive production techniques, understand challenges and make the most of all: channel-, scene- and object-based formats.