AL_INVERSE_DISTANCE_CLAMPED vs AL_LINEAR_DISTANCE_CLAMPED for UI Sound Consistency

Short answer

For a stable minimum audible distance with a fixed AL_REFERENCE_DISTANCE, AL_INVERSE_DISTANCE_CLAMPED is the more forgiving model: inside the reference distance its gain is pinned at 1.0, and beyond it the gain falls off smoothly as ref / (ref + rolloff * (distance - ref)). AL_LINEAR_DISTANCE_CLAMPED ramps linearly to exactly zero at AL_MAX_DISTANCE, which is predictable but guarantees your UI cue eventually goes silent. However, the real fix for UI consistency is usually not the distance model at all — it is setting AL_SOURCE_RELATIVE to AL_TRUE on UI sources, which bypasses distance attenuation entirely.

How each model behaves as rolloff changes

Both models clamp the effective distance to at least AL_REFERENCE_DISTANCE, so nothing gets louder than unity gain up close. The difference shows up beyond that radius:

• Inverse clamped: gain = ref / (ref + rolloff * (d - ref)). Raising AL_ROLLOFF_FACTOR steepens the curve, but gain never reaches zero at any finite distance. The "threshold" where the cue becomes inaudible is soft and depends on your mix level.
• Linear clamped: gain = 1 - rolloff * (d - ref) / (max - ref). Raising rolloff moves the zero point inward; at rolloff 1.0 the sound hits exactly zero at AL_MAX_DISTANCE. The cutoff distance is hard and easy to compute, which is useful for culling but bad for cues that must always be heard.

So with a constant reference distance, the inverse model gives the more stable minimum audible distance, because its tail never fully dies. The linear model gives a more stable maximum distance — the opposite guarantee.

Can per-source AL_REFERENCE_DISTANCE alone preserve both?

Partially, but it is fragile. Since AL_DISTANCE_MODEL is global, you can give UI sources a very large reference distance (so they sit permanently inside the unity-gain zone) while world sources keep small values. Two problems: under the linear model, gain still ramps down once distance exceeds the reference, and a large reference distance interacts with AL_MAX_DISTANCE in ways that are easy to misconfigure — setting max equal to reference is a degenerate case with implementation-dependent behavior. Any future change to world scale or listener motion re-breaks the tuning.

The robust pattern

1. For pure UI cues, set alSourcei(src, AL_SOURCE_RELATIVE, AL_TRUE) and position (0, 0, 0). Distance attenuation is skipped; volume depends only on AL_GAIN.
2. If sources must stay positional (e.g., cues anchored to in-world widgets), keep the global model at AL_INVERSE_DISTANCE_CLAMPED and set AL_ROLLOFF_FACTOR to 0 on UI sources — this disables attenuation per source without touching world sounds.
3. Reserve AL_LINEAR_DISTANCE_CLAMPED for cases where you genuinely want a hard silence cutoff, such as culling offscreen emitters.

Likely cause of your inconsistency, and how to confirm

The most common cause is UI sources left in the default absolute mode with the default reference distance of 1.0 — small listener movements then produce large gain swings under the inverse model. To verify, temporarily set rolloff to 0 on the UI source: if volume becomes constant, distance attenuation was the culprit. Also check that nothing else is modulating AL_GAIN, since per-source gain multiplies with the distance-model gain. Finally, test on your actual backend (e.g., OpenAL Soft); clamping edge cases have historically differed between implementations.