Professional sound,
without the
overhead.

The Recording Setup

A Neumann KU-100 dummy head microphone was positioned on an elevated running track above a gymnasium floor — roughly centered above the court, looking down at a live volleyball practice. The elevated position placed the primary sound sources below the mic plane, which has direct implications for the spatial design work that followed.

The KU-100's capsule geometry and pinnae are designed to replicate human head-related transfer functions, making it well-suited for capturing environments where lateral localization is meaningful. A volleyball court is a particularly effective demonstration: the ball moves discretely left and right across the court, and on headphone playback the binaural image tracks that movement clearly with no visual information required. Inter-aural time and level differences do the work.

The recorded scene included player voices, sneaker impacts on hardwood, whistle cues, ball strikes, crowd bleed from an adjacent court separated by a heavy acoustic curtain, and low-level ambient music from elsewhere in the building.

Spatial Sound Design: Working Without an Impulse Response

The post-production challenge was integrating virtual sound objects into the recorded space without an impulse response of the gym itself. Without a measured IR, there's no ground truth for the room's reverb time, pre-delay, or frequency-dependent decay — all of which have to be approximated by ear against the recorded material.

Large gymnasium acoustics have a recognizable character: long decay times, prominent early reflections off hard parallel surfaces, and significant high-frequency rolloff at distance due to air absorption. Because this acoustic environment is familiar to most listeners, placement errors are perceptible even without critical listening training.

The processing chain for each virtual object used two plugins in sequence in Pro Tools: Space (convolution reverb) for acoustic environment matching and distance simulation, followed by dearVR Micro for binaural spatialization with azimuth and elevation control. No two source samples shared identical processing — each had its own recorded acoustic signature that had to be addressed before spatial placement.

Specific Placement Decisions

Crowd bleed through the curtain: Target position 14° azimuth, -27° elevation. The negative elevation reflects the recording geometry — the mic was above the floor. The Buiksloterkerk IR (a large reverberant church space) was used for the convolution stage to simulate acoustic distance and diffusion. The choice of a dissimilar IR is intentional: convolution reverb here isn't meant to accurately replicate the gym — it's meant to create a sense of enclosure and distance that allows the sound to sit behind the curtain rather than float in the immediate space.

Whistle and buzzer: Placed at the same azimuth but different elevations. The buzzer received positive elevation — buzzers in gymnasium environments are typically ceiling-mounted, and the elevation cue contributes to placement plausibility. The whistle, originating from floor level, was given negative elevation to match the recording geometry.

On Sample Matching

Two crowd samples were used, sourced from different libraries with different recorded acoustic characters. Applying the same convolution IR to both would have introduced detectable processing artifacts — the spectral coloration of an IR applied to already-reverberant material compounds in ways that become audible on careful listening. Each was treated with a different IR chosen to complement rather than conflict with its existing acoustic character, then spatialized independently. This per-element treatment is standard practice in film and television sound, where the editor is almost never working with an IR of the original recording location.

Why not convolve a mono source with per-channel HRIRs?

The intuitive approach — take a mono signal, duplicate it, and convolve each copy with an HRIR corresponding to a target speaker position — doesn't work well, and it's worth understanding why. The spatial information in a multichannel mix lives in the differences between channels. A proper 5.1 mix has discrete content at each speaker position: dialogue anchored to center, effects panned across the field, surround content in the rears. Applying different HRIRs to copies of the same mono signal doesn't replicate that — it produces a single source with multiple colored versions of itself summed together, which tends to sound phasey and spatially unconvincing. The binauraliser needs real per-channel content to work with.

Signal chain: binaural render

Starting with a native 5.1 source file, the routing in Pro Tools: 5.1 audio track → 5.1 aux (processing) → stereo aux (binaural output). Plugin on the stereo aux: Sparta Binauraliser v1.7.0 with SADIE database subject H6 SOFA as the HRTF. Speaker positions set to 30°, -30°, 0°, 110°, -110° (standard 5.1 layout). Interpolation mode: Triangular (PS). Output captured to a stereo audio track via internal record, then bounced.

The SADIE H6 SOFA is a well-validated general-purpose HRTF, widely used in academic and research contexts and compatible with any binauraliser that accepts SOFA-format files. Sparta's Triangular PS interpolation handles panning between measured HRIR positions cleanly without introducing artifacts at boundaries.

Signal chain: stereo downmix

The same 5.1 source was routed to a separate aux dedicated to downmix processing. Plugin: dearVR PRO 2, output format set to 2.0 Stereo (2CH). A small amount of reverb was retained — fully dry, certain transients in the source material produced audible artifacts, likely a characteristic of the original recording rather than the plugin. Keeping the binaural render chain and the downmix chain separate allows independent A/B comparison during review.

On HRTF selection

HRTF choice matters more than is often assumed. Individual HRTFs vary significantly in elevation cues and front/back localization characteristics — a set that produces convincing externalization for one listener may collapse to in-head localization for another. For delivery contexts where individualized HRTFs aren't practical, a well-validated non-individual set like SADIE H6 or KEMAR is a reasonable starting point. For critical listening or immersive content intended for wide release, testing across multiple HRTF sets is worthwhile.