Technical Documentation

Analysis Modes

Six spectral analysis engines — from industry-standard FFT to super-resolution Superlet — each optimized for different measurement scenarios

Why Multiple Analysis Modes?

Every spectral analysis method faces a fundamental trade-off: frequency resolution vs. time resolution. A single FFT can give you fine frequency detail, but only by sacrificing responsiveness. AudioMap solves this by offering six distinct analysis engines, each with a different approach to the time-frequency trade-off.

Choose the mode that best fits your measurement scenario — whether you need the speed of a traditional FFT, the multi-resolution detail of MRA, the musical pitch resolution of CQT, the smoothness of CWT, or the surgical precision of Superlet. You can also sharpen any compatible mode with STFT Reassignment.

Fast Fourier Transform

Traditional single-FFT analysis

The industry-standard approach to spectral analysis. A single FFT of configurable size transforms time-domain audio into frequency-domain data. Larger FFT sizes give finer frequency resolution at the cost of slower time response.

How It Works

Audio samples are accumulated in an overlapping ring buffer

A Hann window is applied to reduce spectral leakage

75% overlap (hop = FFT size / 4) provides smooth temporal updates

FFT is computed using vDSP (Apple Accelerate) for sizes up to 4096, or pocketfft for larger sizes

Magnitude and phase are extracted from the complex spectrum

Best For

General-purpose frequency analysis
Standard measurement workflows
When uniform frequency resolution is acceptable
Compatibility with traditional analysis tools

Considerations

Fixed frequency resolution across all frequencies
Low-frequency resolution requires very large FFT sizes
Large FFTs introduce latency (e.g., 131K = 2.7s at 48kHz)
Trade-off between frequency and time resolution

Technical Specifications

FFT Sizes128 to 131,072

11 sizes (powers of 2)

WindowHann

Optimal spectral leakage control

Overlap75%

Hop = FFT size / 4

Default Size4096

~1 Hz resolution at 48 kHz

BackendHybridFFT

vDSP + pocketfft fallback

Max Buffer2,097,152 samples

Supports up to 256K FFT

FFT Size Guide

Resolution and latency at 48 kHz sample rate

Size	Resolution	Latency	Use Case
1024	~47 Hz	~21 ms	Fast response, live monitoring
4096	~12 Hz	~85 ms	General purpose (default)
16384	~3 Hz	~341 ms	Detailed frequency analysis
65536	~0.7 Hz	~1.4 s	Sub-bass resolution
131072	~0.4 Hz	~2.7 s	Maximum resolution

Window Functions

Applied to the time-domain signal before FFT to control spectral leakage — available for FFT and MRA modes

Hann

Default

The default and recommended choice. Excellent spectral leakage suppression with good frequency resolution. Standard for acoustic measurements.

Hamming

Similar to Hann but with slightly less sidelobe suppression. Minimizes the nearest sidelobe at the cost of slower far-sidelobe decay.

Blackman-Harris

Low Leakage

Very low sidelobes (-92 dB) for maximum spectral leakage rejection. Wider main lobe means slightly reduced frequency resolution. Use when dynamic range matters most.

Rectangular

No Window

No windowing (raw FFT). Best frequency resolution but severe spectral leakage. Only useful for signals that are exactly periodic within the FFT window.

Flat-Top

Calibration

Optimized for amplitude accuracy at the expense of frequency resolution. Ideal for calibration and precise level measurements of known tones.

For most acoustic measurement work, Hann is the correct choice. Switch to Blackman-Harris when measuring signals with large dynamic range differences between frequency components, or Flat-Top when precise amplitude calibration is needed.

Temporal Averaging & Smoothing

Configure how measurements are averaged over time and smoothed across frequency — applies to all analysis modes

Temporal Averaging

None

No averaging — raw per-frame spectrum. Maximum responsiveness, most noise.

FIFO

Default

Moving average over a circular buffer of recent frames. Equal weight to all frames in the window.

Window sizes: 1, 2, 4, 8, 16 (default), 32, 64, 128

LPF (Low-Pass Filter)

5th-order Bessel filter with maximally flat group delay. Smooth, continuous averaging without window boundaries.

Cutoff: 0.1 – 10.0 Hz (default: 1.0 Hz)

Averaging Domains

Complex— Phase-sensitive; can reduce magnitude with phase variation

Mag+Phase— Average magnitude and phase separately; preserves magnitude

Power ✦— RMS-style averaging (default). Industry standard, always positive

Frequency Smoothing

Option	Bandwidth
None	No smoothing
1/48 octave	Ultra-fine (0.02 oct)
1/24 octave	Very fine (0.04 oct)
1/12 octaveDefault	Default — recommended
1/6 octave	Moderate smoothing
1/3 octave	Standard measurement
1 octave	Heavy smoothing
Variable	Frequency-dependent
Psychoacoustic	Perception-matched

Variable Smoothing

No smoothing below 100 Hz, ramps to 1/6 octave at 1 kHz, fixed 1/6 octave above. Preserves low-frequency detail while taming high-frequency noise.

Psychoacoustic Smoothing

1/3 octave below 100 Hz, interpolates to 1/6 octave at 1 kHz, fixed 1/6 above. Uses cubic mean weighting to match human frequency perception.

Phase Smoothing

Available in Measure mode (transfer function). Smooths unit vectors (cos/sin) independently then recovers phase via atan2 — prevents discontinuities at ±180° boundaries. Coherence-weighted for accuracy.

Mode Comparison

Choose the right analysis mode for your workflow

Feature	FFT	MRA	CQT	CWT	Superlet
LF Resolution	Size-dependent	Excellent	Excellent	Good	Excellent
HF Time Response	Good	Excellent	Good	Good	Good
CPU Usage	Low	Moderate	Moderate	High	Very High
Transfer Function	Full	Full	H1	H1	H1
Reassignment	Yes	Yes	Yes	No	No
Frequency Scale	Linear	Linear	Log (native)	Log	Log
Key Parameter	FFT Size	Stage FFT Size	Bins/Octave	Bandwidth (σ)	Max Order

Back to Support Setup Guide