# Receiver Placement Eval

> Tests an algorithm's ability to optimally place a receiver in 3D indoor environments to maximize speech intelligibility, measured by the Speech Transmission Index (STI). It evaluates how well the optimization handles complex acoustic properties like reverberation and noise across different scene geometries. Use when the user wants to benchmark on Office, Berlin, Suburban 3D scenes, or asks about evaluating this task. Reports STI.

- Skill: `qhjqhj00/receiver-placement-eval` (Agent Skill)
- Install (CLI): `npx skillmds add qhjqhj00/receiver-placement-eval`
- Raw SKILL.md: https://api.skillmd.com/api/skills/qhjqhj00/receiver-placement-eval/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: qhjqhj00 (https://skillmd.com/u/qhjqhj00)
- Updated: 2026-09-08
- Page: https://skillmd.com/skills/qhjqhj00/receiver-placement-eval

---


# receiver-placement-eval

> Receiver Placement for Speech Enhancement using Sound Propagation Optimization — Morales et al. (2018) (arXiv:1805.11533, 2018)

## What this evaluates

Tests an algorithm's ability to optimally place a receiver in 3D indoor environments to maximize speech intelligibility, measured by the Speech Transmission Index (STI). It evaluates how well the optimization handles complex acoustic properties like reverberation and noise across different scene geometries.

## Datasets

- **Office, Berlin, Suburban 3D scenes** — total ?; splits: test (-1)

## Metrics

- `STI` **(primary)** — range: [0, 1]
  - Speech Transmission Index, a perceptual metric of speech intelligibility computed from Room Impulse Responses (RIRs). Values range from 0 to 1, with higher values indicating better intelligibility. The paper maps STI to qualitative ratings (e.g., >0.76 is A+, <0.36 is U) and uses a Just Noticeable Difference (JND) threshold of 0.03.

## Input / output format

**Input**: 3D indoor scene mesh (CAD model) with annotated material absorption properties, fixed noise source locations, and a discrete search space of candidate receiver positions.

**Output**: Optimized 3D coordinates for the receiver placement that maximizes STI.

## Scoring recipe

```python
def score(optimized_pos, scene_mesh, source_pos):
    rir = simulate_rir(scene_mesh, source_pos, optimized_pos)
    sti = compute_sti_from_rir(rir)
    return sti
# The protocol evaluates placement by computing STI from the simulated RIR.
# Higher STI indicates better speech intelligibility. Improvement is delta_sti = sti_after - sti_before.
```

## Common pitfalls

- The search space is limited to a discrete set of samples (e.g., 879 for Office), so the reported STI may not reflect the true global maximum.
- STI is highly sensitive to material absorption and scene geometry; results do not generalize to environments with unmodeled acoustic treatments or different source-receiver configurations.
- Empirical STI estimation models based solely on room volume ignore spatial geometry and can deviate significantly from accurate propagation-based calculations.

## Evidence (verbatim from paper)

> We tested our receiver placement optimization on three workplace and residential scenes. Our method can accurately compute the STI on complex indoor scenes. The optimization was computed on a desktop machine, using 8 threads. Our benchmark scenes included Office, a multi-room workplace with conference rooms; Suburban, the ground floor of a multi-story house; and Berlin, a small apartment with two connected rooms.

## Citation

```bibtex
@misc{morales2018receiver,
  title={Receiver Placement for Speech Enhancement using Sound Propagation Optimization},
  author={Morales et al. (2018)},
  year={2018},
  note={arXiv:1805.11533}
}
```

- arXiv: 1805.11533