TL;DR
For decades, audio engineers have relied on "close miking" rules of thumb to capture clean sound in noisy environments. This paper provides the first systematic, quantitative verification of these practices by treating close miking as a "mechanical source separation" technique and measuring its performance using the Signal-to-Interference Ratio (SIR).
What Problem Does This Paper Address?
In live recording or multi-instrument sessions, engineers face the challenge of "leakage"—unwanted sound from ambient noise or other instruments bleeding into a microphone. The standard solution is close miking: placing the microphone very near the target source (typically between 0.03m and 1m).
While this practice is ubiquitous, it has historically been guided by empirical intuition rather than hard data. The authors sought to bridge this gap by asking: Can we objectively verify the effectiveness of close miking? By applying metrics typically used to evaluate digital source separation algorithms, they aimed to define the optimal distances, angles, and microphone types for maximizing signal clarity.
Methods
The researchers framed close miking as a physical version of a sound source separation (SSS) task. To evaluate it, they used the Signal-to-Interference Ratio (SIR), a metric that quantifies the energy ratio between the target signal and the unwanted interference.
The core of their analytical approach relies on the disjointness orthogonality assumption. This assumes that in the time-frequency domain (Short-Time Fourier Transform or STFT), the target source and the noise source do not significantly overlap. By applying time-frequency masking, they could isolate how much "interference" from the noise source remained in the recording of the target source across various physical configurations.
The experimental setup (illustrated in Figure 1) involved a controlled recording environment on a reverberant theater stage. They tested:
- Two Microphone Types: A cardioid (directional) Shure SM57 and an omni-directional Behringer ECM8000.
- Three Angles: 0°, 30°, and 45° relative to the target source's central axis.
- Twelve Distances: Ranging from 3cm to 1 meter.
- Varying Sound Pressure Levels (SPL): Different volume ratios between the target guitar amp and a pink noise generator.
Datasets
The authors generated a custom dataset for this study, as no existing benchmarks provided the necessary controlled physical variations.
- Target Source: A clean, repetitive electric guitar riff.
- Interference: A pink noise generator to simulate ambient acoustic noise.
- Environment: A municipal theater stage in Greece. The room's acoustics were characterized using T20, T30, and Early Decay Time (EDT) measurements (shown in Figure 2) to ensure the results were applicable to real-world, reverberant spaces.
Key Insights and Ideas
- Mechanical Separation: The paper successfully validates the idea that physical microphone placement acts as a "mechanical" precursor to digital signal processing, effectively performing the first stage of source separation.
- The Directionality Multiplier: Directional (cardioid) microphones don't just sound different; they provide nearly double the SIR of omni-directional microphones in close-miking scenarios.
- The Power of the Angle: One of the most significant findings is that angling a cardioid microphone (30° to 45°) away from the noise source doesn't just change the frequency response—it triples the SIR compared to a straight-on placement.
- Non-Linear Degradation: SIR doesn't always drop linearly with distance. In some configurations, the researchers observed "dips" in performance at specific distances (e.g., between 20-40cm for cardioid) followed by slight recoveries, likely due to the complex interaction of room reflections and microphone polar patterns.
Key Results
- Optimal Distance: For a standard cardioid microphone pointed directly at the source, the peak separation performance occurs at 5 cm.
- Angled Performance: When the cardioid microphone is placed at an angle of 30° or 45°, the optimal distance shifts slightly further back to approximately 12–14 cm, but the resulting SIR is three times higher than the straight-on 0° placement.
- Cardioid vs. Omni: The cardioid microphone consistently outperformed the omni-directional one across all distances. As seen in Figure 3 (Omni) vs Figure 4 (Cardioid), the cardioid's SIR values are significantly higher, confirming its superior "mechanical separation" capabilities.
- SPL Correlation: As expected, higher source SPL relative to the noise SPL leads to a higher SIR, confirming that the technique scales predictably with volume.
Limitations and Future Work
- Timbre and Proximity Effect: The study focused purely on separation (SIR). It did not account for the "Proximity Effect" (the boost in bass frequencies when a directional mic is close to a source), which is a major factor in an engineer's subjective choice of distance.
- Microphone Height: The researchers noted that the height of the microphone relative to the source was not varied in this study and could be a subject for future investigation.
- Subjective Quality: While the SIR metric is objective, the authors suggest that future work should include subjective listening tests to see if the "optimum" separation distances also correlate with what listeners perceive as the "best" sounding recording.
Summary updated