Question 1

What FFT size should I use for general audio analysis?

Accepted Answer

For general audio analysis at 48 kHz sample rate, 4096 points is a good starting point. It provides 11.7 Hz frequency resolution with approximately 85 ms time blocks, giving about 12 updates per second — responsive enough for real-time monitoring. If you need to see finer frequency detail (identifying specific tones, resolving room modes), increase to 8192 or 16384. If you need faster response for tracking transient sounds, decrease to 2048 or 1024. For live sound system tuning, 4096 or 8192 is standard. For noise analysis requiring identification of discrete tones, use 8192 or higher. For music monitoring where visual responsiveness matters, 2048 is sufficient. SonaVyx lets you change FFT size during measurement, so you can start with 4096 and increase as needed.

Question 2

What type of averaging should I use?

Accepted Answer

SonaVyx offers four averaging modes. No averaging shows the instantaneous spectrum, which is noisy but captures transients. Linear averaging takes the arithmetic mean of N spectra, giving equal weight to all frames — useful for measuring steady-state noise where you want the time-average. Exponential averaging weights recent frames more heavily, providing a smoothed but responsive display — best for live monitoring where conditions change slowly. Peak hold retains the maximum value at each frequency across all frames — useful for identifying the worst-case peak levels. For system tuning, use exponential averaging with 4-8 frames. For compliance measurements (Leq), use linear averaging over the required time period. For feedback detection, use no averaging or fast exponential to catch rapid changes.

Question 3

What is spectral smoothing and how much should I apply?

Accepted Answer

Spectral smoothing averages the magnitude across a bandwidth proportional to frequency (fractional octave). This reduces the visual complexity of the display by merging narrow-band features. Smoothing levels: 1/24 octave (minimal smoothing, shows nearly raw FFT detail), 1/12 octave (resolves room modes), 1/6 octave (standard for system measurement, resolves broad features), 1/3 octave (standard for room acoustics, matches human frequency discrimination), 1/1 octave (very smooth, for broad tonal overview). For system EQ decisions, 1/6 octave shows what parametric EQ can realistically correct. For noise analysis, 1/3 octave matches standard octave band filters. For detailed investigation of specific problems, use 1/12 or less. SonaVyx applies smoothing in real time, so you can adjust while viewing to find the most informative level.

Question 4

What overlap percentage should I use for FFT analysis?

Accepted Answer

Overlap determines how much each FFT frame shares samples with the previous frame. Zero overlap means completely new data each frame, giving the slowest update rate but independent frames. 50% overlap (the standard) means half the data is shared, doubling the update rate. 75% overlap quadruples the rate but with more redundancy between frames. Higher overlap gives smoother visual updates and faster tracking of changes, but does not add new information — it interpolates between frames. For transfer function measurement using Welch averaging, 50% overlap is standard and matches the standard statistical analysis assumptions. For real-time display, 75% overlap gives a smoother visual experience. SonaVyx defaults to 50% overlap, which is appropriate for both analysis and display.

Question 5

How does the dB range setting affect what I see?

Accepted Answer

The dB range sets the vertical scale of the spectrum display. A wider range (e.g., -100 to 0 dB) shows more of the signal, from peak levels down to the noise floor, but compresses the visual resolution of small level differences. A narrower range (e.g., -40 to +10 dB) zooms in on the signal of interest, making small differences more visible but potentially clipping peaks or hiding the noise floor. For general monitoring, -80 to 0 dB works well. For system tuning where you care about ±3 dB variations, narrow the range to about 40 dB centered on your signal level. For noise analysis, set the bottom of the range near the noise floor. SonaVyx magnitude plot defaults to -40 dB to +20 dB (60 dB range) and supports scroll-to-zoom to dynamically adjust the visible range.

Question 6

When should I use a Hann window versus a Blackman-Harris window?

Accepted Answer

Hann (also called Hanning) has a narrow main lobe (good frequency resolution) but moderate sidelobe rejection (-31 dB). Use Hann when resolving closely spaced frequencies is more important than measuring weak signals near strong ones. Blackman-Harris has a wider main lobe (slightly less frequency resolution) but excellent sidelobe rejection (-92 dB). Use it when you need to see low-level components near a strong tone, such as measuring distortion harmonics 60 dB below the fundamental, or detecting a quiet noise source in the presence of a loud one. For general audio measurement, Blackman-Harris is safer because it prevents leakage artifacts from misleading you. SonaVyx defaults to Blackman-Harris. Switch to Hann when you need the finest frequency resolution, such as resolving closely spaced room modes.

Question 7

What is zero-padding and should I use it?

Accepted Answer

Zero-padding appends zeros to the time-domain data block before computing the FFT, increasing the FFT size without adding new information. This interpolates between frequency bins, making the spectrum display smoother and peaks easier to identify visually. It does not improve actual frequency resolution (which is determined by the data length, not the FFT size). Zero-padding a 4096-point block to 8192 points gives 8192 output bins but with the same 11.7 Hz resolution as a 4096-point FFT. Use zero-padding when you want smoother visual display without the slower update rate of a larger actual FFT. SonaVyx does not currently expose a zero-padding control because the combination of large FFT sizes and smoothing provides equivalent visual clarity with genuine resolution improvement.

Question 8

How do I optimize settings for identifying low-frequency room modes?

Accepted Answer

Room modes in a typical room (3-8 m dimensions) occur between 20 and 300 Hz, with the fundamental axial modes at f = c/(2L). To resolve these: use a large FFT size (16384 or 32768) for frequency resolution of 2.9 or 1.5 Hz. Use minimal smoothing (1/24 octave or none) so narrow peaks are visible. Use linear or exponential averaging (8-16 averages) to reduce noise variance. Excite the room with broadband low-frequency content (pink noise through a subwoofer or the room test signal). Place the microphone in a corner where all modes are maximally excited. The spectrum will show a series of peaks at the mode frequencies. Compare observed peaks with calculated mode frequencies (f = c × sqrt((nx/2Lx)² + (ny/2Ly)² + (nz/2Lz)²)) to identify axial, tangential, and oblique modes. SonaVyx room scan calculates expected mode frequencies for comparison.

Spectrum Analyzer Settings: FFT Size, Window, Averaging, and Smoothing Guide

Frequently Asked Questions