Audio Metering

Accurate metering is critical for mixing, mastering, and broadcast compliance. Different meter types measure different properties — peak meters catch transients, RMS/VU meters approximate perceived loudness, and EBU R128 defines the modern loudness standard. REVITHION STUDIO needs comprehensive metering across all of these standards to ensure professional output at every stage.

Peak Meter

Sample-peak metering takes the absolute value of each sample, tracking the maximum per buffer. Ballistics use instant attack with a slow release fall rate of 1–2 dB/sec. Display in dBFS (dB relative to full scale): dBFS = 20 * log10(|sample|).

float peakLevel = 0.0f;
for (int i = 0; i < numSamples; ++i)
    peakLevel = std::max(peakLevel, std::abs(buffer[i]));
float peakDb = 20.0f * std::log10(peakLevel + 1e-6f);

Peak hold retains the maximum value for approximately 2 seconds, then releases at the standard fall rate. Track peakHoldValue and peakHoldTimer; reset the timer on each new peak, decrement per callback, and begin release when the timer expires.

RMS Meter

RMS (Root Mean Square) measures average signal power over a window, which is more perceptually relevant than peak. A 300 ms window is typical for general use; per-buffer RMS works for visualization. Convert to dB for display.

float sumSquares = 0.0f;
for (int i = 0; i < numSamples; ++i)
    sumSquares += buffer[i] * buffer[i];
float rms = std::sqrt(sumSquares / static_cast<float>(numSamples));
float rmsDb = 20.0f * std::log10(rms + 1e-6f);

VU Meter

The VU (Volume Unit) meter uses approximately 300 ms integration time, reaching 99% of a steady-state value in that window. It roughly corresponds to perceived loudness for sustained signals. Historically 0 VU equals +4 dBu at professional reference level. In digital, calibrate 0 VU to −18 dBFS (EBU convention) or −20 dBFS (SMPTE convention). Implement with a single-pole IIR lowpass on the rectified signal to model the ballistic response.

EBU R128 / ITU-R BS.1770 Loudness

K-Weighting Filter

Two cascaded biquad filters applied before loudness measurement. The first is a high shelf boosting +3.9999 dB at high frequencies, modeling the head-related diffuse-field response. The second is an RLB high-pass filter removing energy below ~60 Hz. Coefficients are sample-rate dependent — recalculate when the rate changes.

// JUCE IIR biquad setup for K-weighting (48 kHz coefficients shown)
auto highShelf = juce::dsp::IIR::Coefficients<float>::makeHighShelf(48000.0, 1681.97, 0.7071f, 3.9999f);
auto highPass  = juce::dsp::IIR::Coefficients<float>::makeHighPass(48000.0, 38.1355, 0.5003f);

Loudness Measurements

• Momentary Loudness (M): 400 ms sliding window over K-weighted signal. Updated every 100 ms.
• Short-term Loudness (S): 3-second sliding window over K-weighted signal. Updated every 100 ms.
• Integrated Loudness (I): Entire program duration, gated. Absolute gate discards blocks below −70 LUFS. Relative gate computes Γ_a (mean of blocks passing absolute gate), then discards blocks below Γ_a − 10 LU.
• Loudness Range (LRA): Statistical distribution of short-term loudness between the 10th and 95th percentiles after gating.
• True Peak (TP): Oversampled peak detection (minimum 4× per ITU-R BS.1770) to catch inter-sample peaks that exceed 0 dBFS after DAC reconstruction.
• Units: LUFS (Loudness Units Full Scale) for absolute, LU for relative measurements.

Targets

Context	Integrated Loudness	True Peak Max
Spotify / YouTube	−14 LUFS	−1 dBTP
Apple Music	−16 LUFS	−2 dBTP
EBU R128 (EU broadcast)	−23 LUFS ± 0.5 LU	−1 dBTP
ATSC A/85 (US broadcast)	−24 LUFS ± 2 LU	−2 dBTP

K-System Metering (Bob Katz)

RMS-based metering with peak indicators and a shifted 0-point: K-20 (0 = −20 dBFS) for wide dynamic range material like classical and jazz; K-14 (0 = −14 dBFS) for pop and rock; K-12 (0 = −12 dBFS) for broadcast. The system encourages mixing to a calibrated monitoring level rather than maximizing peak level.

Spectral Analyzer

Apply a windowed FFT (Hann window, 2048 or 4096 samples) to compute the magnitude spectrum. Display on a logarithmic frequency axis (20 Hz–20 kHz) to match octave perception, with a dB scale on the Y axis (typically −90 to 0 dBFS). Smooth magnitudes over time with exponential averaging for a stable display.

// Exponential smoothing per FFT bin for display stability
smoothedMag[bin] = smoothCoeff * smoothedMag[bin]
                 + (1.0f - smoothCoeff) * currentMag[bin];

Correlation Meter

Measures the phase relationship between L and R channels. +1.0 means perfectly correlated (fully mono compatible), 0.0 means uncorrelated (wide stereo), and −1.0 means perfectly out of phase (mono cancellation — a serious problem). Implement as a running normalized cross-correlation.

sumLR += left * right;
sumLL += left * left;
sumRR += right * right;
float correlation = sumLR / (std::sqrt(sumLL * sumRR) + 1e-6f);

Stereo Scope (Goniometer / Lissajous)

An X-Y display of the stereo signal rotated 45° into the M/S domain. Vertical axis represents mid (mono sum), horizontal axis represents side (difference). Plot rotated coordinates: x = (R - L) / sqrt(2), y = (L + R) / sqrt(2). Useful for visualizing stereo width, balance, and phase issues at a glance.

Phase Meter

A real-time visual representation of the correlation value, typically rendered as a needle or bar oscillating between −1 and +1. Often combined with the goniometer display. Values persistently below 0 indicate phase problems that will cause mono fold-down issues on broadcast systems.

Thread Safety for Metering

Meter values are computed on the audio thread but displayed on the GUI thread. Never use a mutex or lock on the audio thread. Use std::atomic<float> with relaxed memory ordering — sufficient for single-producer single-consumer scalar metering data.

std::atomic<float> peakLevelL{0.0f};
std::atomic<float> peakLevelR{0.0f};

// Audio thread: store with relaxed ordering (no fence needed)
peakLevelL.store(newPeak, std::memory_order_relaxed);

// GUI thread: load with relaxed ordering
float displayPeak = peakLevelL.load(std::memory_order_relaxed);

Anti-Patterns

• ❌ Don't use sample-peak as a loudness meter — peak ≠ loudness
• ❌ Don't skip true-peak metering for mastering — inter-sample peaks cause DAC clipping
• ❌ Don't use mutex/lock to transfer meter values from the audio to GUI thread
• ❌ Don't forget K-weighting for loudness measurement — unweighted RMS is not perceptual loudness
• ❌ Don't display raw FFT bins on a linear frequency axis — use logarithmic for musical relevance

Checklist

• Peak meter includes true-peak (oversampled) option
• RMS window length configurable (50 ms to 1000 ms)
• EBU R128 integrated loudness uses both absolute and relative gates
• K-weighting filters recalculated for current sample rate
• Meter values passed via std::atomic<float> — no locks on audio thread
• Spectral analyzer uses windowed FFT with logarithmic frequency axis
• Correlation meter active for mono compatibility checking