Block-diagram of the short-window, fast processing approach (DynBU_fast) is shown in the top panel a. The left and right ear signals are first bandpass filtered using a Gammatone filter bank parametrized along the Bark scale. The time signal of each filter is windowed with 12 ms overlapping Hanning windows, resulting in an effective window length of 6 ms. The interaural cross-correlation of the interferer (ρ_i) as well as the interaural phase difference of target and interferer (ϕ_t and ϕ_i) are extracted for each filter and each time window to calculate binaural unmasking according to formula (1). A 225 ms exponential decay filter is subsequently used to account for sluggishness of binaural processing. Binaural unmasking and the better-ear SNR are added for each frequency band and time frame, followed by selecting the maximum of the binaural benefit across frequency bands per time frame. The binaural benefit for signal detection is estimated by selecting the maximum binaural benefit over time. The DynBU_slow approach (see Sect. 4.3) differs from DynBU_fast only by using a 225 ms window directly after the Gammatone filterbank to derive the BMLDs without integration afterwards. The DynBU_slow approach is shown in lower panel b of the figure.