Audio DSP Glossary

Attack time controls the exponential coefficient for the envelope follower. Formula: coeff = exp(-1 / (time * sampleRate)). At 10ms attack (48kHz): coeff = exp(-1 / (0.01 * 48000)) ≈ 0.9998. Higher coefficient = slower attack.

In TabDSP: Attack applies when gain reduction is increasing (signal going over threshold). Fast attack (1-5ms) catches transients, slow attack (50-100ms) lets transients through.

Achieved via lookahead buffer: the limiter scans ahead for upcoming peaks and applies gain reduction BEFORE they reach the output. Instant attack (0 samples) is possible because peaks are predicted. Zero overshoot guaranteed.

See: True Peak Limiter

TabDSP uses crossfade bypass with a complementary wet/dry gain pair: the dry path fades to 1.0 while the wet (processed) path fades to 0.0 over 30ms. This prevents clicks and pops. Bypassed modules show orange routing arcs in the UI.

See: Signal Flow

Ceiling is enforced via instant attack in the lookahead limiter. When a peak above ceiling is detected in the lookahead buffer, gain reduction is applied immediately: gain = ceiling / peak. True peak detection (4x oversampling) prevents inter-sample peaks from exceeding ceiling.

See: True Peak Limiter

TabDSP compressor uses RMS envelope follower with soft knee. Gain reduction formula: gr = (inputDb - threshold) * (1 - 1/ratio) above threshold. Knee creates smooth transition region. Attack/release use exponential smoothing.

See: Single-Band Compressor

TabDSP uses native AnalyserNode FFT (8192-point) with constant-Q power summation: 512 log-spaced bins, Q=12. Linear FFT bins are remapped into logarithmic frequency bands by weighting overlapping bins and summing their power. Blackman window with coherent gain compensation (+7.535 dB) ensures accurate absolute levels.

Advantage: Constant relative bandwidth matches musical perception, while leveraging the browser's native C++ FFT engine for zero JavaScript CPU overhead.

TabDSP uses Linkwitz-Riley 24dB/oct (LR4) crossovers. LR4 = two cascaded 2nd-order Butterworth filters (Q = 0.707). LP and HP both at -6dB at crossover frequency, sum to unity (0dB) with phase coherence. This ensures transparent band splitting.

See: Multiband Compressor

Formula: dB = 20 * log10(amplitude). Conversions: +6dB = 2x amplitude, +12dB = 4x, +20dB = 10x. Reverse: amplitude = 10^(dB/20). In digital audio, dBFS (Full Scale) means 0 dBFS is the maximum possible level before clipping.

In 24-bit audio, 0 dBFS = sample value of ±8,388,607 (2^23 - 1). Any value above this wraps around, creating harsh distortion. dBFS measures sample peaks only—doesn't account for inter-sample peaks during DAC reconstruction (see dBTP).

Measured via 4x oversampling: audio is interpolated to 4x sample rate using polyphase FIR filters, then peaks are detected in the upsampled signal. ITU-R BS.1770-4 specifies the filter coefficients. True peak can be 0.5-3 dB higher than sample peak due to inter-sample reconstruction.

See: True Peak Limiter

Exponential smoothing implementation: If target > current, use attack coefficient. If target < current, use release coefficient. Formula: current += (target - current) * (1 - coeff). Coefficient = exp(-1 / (time * sampleRate)).

FIR filters convolve input with pre-computed coefficients. TabDSP uses FFT-based overlap-add convolution (1024-point radix-2 Cooley-Tukey FFT) instead of sample-by-sample multiplication, achieving ~4x fewer operations. Linear phase achieved via symmetric coefficients. Latency = taps/2 samples. TabDSP: 511 taps / 2 = 255.5 samples ≈ 5.3ms at 48kHz, plus 5.3ms lookahead = ~11ms total.

See: Linear Phase EQ

Measured in dB relative to a reference. EQ bands directly shape the frequency response curve. TabDSP displays the resulting curve as a white line overlaid on the spectrum analyzer.

Calculated as: grDb = (inputDb - threshold) * (1 - 1/ratio) for signal above threshold. Example: input = -10 dB, threshold = -18 dB, ratio = 4:1. Over = 8 dB. GR = 8 * (1 - 0.25) = 6 dB. Output = -10 - 6 = -16 dB (before makeup).

Aim for -18 to -12 dBFS RMS at the input stage. Monitor the IN and OUT meters to verify levels stay healthy throughout the chain. The INPUT module knob adjusts the initial level before processing.

With a limiter ceiling of -1.0 dBTP, the limiter provides 1 dB of headroom below true peak clipping. The input gain, compressor threshold, and limiter ceiling all affect available headroom at different points in the chain.

True peak detection via 4x polyphase oversampling catches these invisible peaks. TabDSP's limiter uses ITU-R BS.1770-4 compliant interpolation (ITU 48-tap or Kaiser 128-tap filters) to measure and limit inter-sample peaks. The True Peak hold bar (orange) on the output meter shows the highest detected inter-sample peak.

Knee creates transition region: [threshold - knee/2, threshold + knee/2]. Within this region, compression ratio gradually increases from 1:1 to target ratio using quadratic curve: gr = (over²) / (2 * knee) * (1 - 1/ratio). This prevents harsh "kink" at threshold.

LR4 = two cascaded 2nd-order Butterworth filters (Q = 0.707 each). At crossover frequency, LP and HP both measure -6 dB. When summed: (-6dB)² + (-6dB)² = 0 dB (unity). This is mathematically guaranteed by the Butterworth Q value.

See: Multiband Compressor

Ring buffer implementation: write pointer advances with incoming audio, read pointer lags by lookahead samples. Peak detection scans between write and read pointers. When peak > ceiling, instant gain reduction applied. Latency = lookahead time + processing delay.

See: True Peak Limiter

ITU-R BS.1770-4 algorithm: (1) K-weighting filter (shelves emphasizing mid frequencies), (2) Mean square calculation per channel, (3) Gating (ignore blocks below -70 LUFS absolute, -10 LUFS relative), (4) Logarithmic conversion. Three modes: Momentary (400ms), Short-term (3s), Integrated (whole program).

Formula: LUFS = -0.691 + 10 * log10(sum of weighted power)

The LUFS Stack uses an ITU-R BS.1770-4 compliant worklet running as a parallel tap from the output analysis point (analysis-only, not in the audio signal path). Target Validator supports three modes: I (LUFS-I target), LRA (Loudness Range), and TP (True Peak). The histogram offers two display modes: SCROLL (10-second rolling window) and FIT (full analysis history). Click Analyze to start collecting data; click Reset to clear all accumulated measurements.

Defined in EBU R 128. LRA uses the same gated loudness measurement as Integrated LUFS but examines the statistical distribution of Short-term (3-second) loudness blocks. After absolute gating (-70 LUFS) and relative gating (-20 LU below ungated mean), the 10th and 95th percentile loudness values are computed. LRA = 95th percentile - 10th percentile. Available in TabDSP's LUFS Stack Target Validator when LRA mode is selected.

Applied after gain reduction: output = input * grLinear * makeupLinear. Makeup is linear gain (not dB), converted via: makeupLin = 10^(makeupDb/20). Auto-makeup (not implemented in TabDSP) would set makeup = average GR.

Matrix encoding: Mid = (L + R) / 2, Side = (L - R) / 2. Process Mid and Side independently. Decode: L = Mid + Side, R = Mid - Side. Division by 2 prevents level increase. Mid contains mono/center content, Side contains stereo difference.

See: M/S Processing

Gain reduction formula: grDb = overDb * (1 - 1/ratio). Example: 4:1 ratio, 8 dB over threshold. GR = 8 * (1 - 0.25) = 6 dB. Infinity:1 = brickwall limiting (any amount over threshold becomes 0 dB over).

Release applies when gain reduction is decreasing (signal below threshold). Uses exponential decay: grCurrent += (0 - grCurrent) * (1 - releaseCoeff). Coefficient = exp(-1 / (releaseTime * sampleRate)). Slower release = higher coefficient = more smoothing.

Formula: RMS = sqrt(sum of squared samples / window size). TabDSP uses 10ms window (480 samples at 48kHz). Running sum for efficiency: sum -= oldest²; sum += newest². Periodic recalculation every 10s prevents floating-point drift.

See: Single-Band Compressor

True sidechain: audio is duplicated, one copy filtered for detection, other copy compressed based on filtered detection. Filter does NOT affect output audio. TabDSP: 2nd-order Butterworth HP at 20-500 Hz on detection path only. This prevents low-frequency energy from dominating compression decisions.

See: Single-Band Compressor

Comparison in dB domain: if (inputDb > threshold) { compress }. Gain reduction only applied when signal exceeds threshold. Knee creates smooth transition region around threshold rather than hard cutoff.

Fast attack settings (1-5ms) will catch and reduce transients, resulting in a smoother but potentially less punchy sound. Slow attack (50-100ms) lets transients pass through, preserving punch and presence at the cost of less peak control.

Detected via 4x polyphase FIR oversampling. Audio interpolated to 4x sample rate, creating "virtual samples" at 0.25, 0.5, 0.75 positions between real samples. True peak = max across all real and virtual samples. ITU-R BS.1770-4 specifies filter coefficients. Can be 0.5-3 dB higher than sample peak.

See: True Peak Limiter