Skip to main content
Fig. 11 | EURASIP Journal on Audio, Speech, and Music Processing

Fig. 11

From: DOA-guided source separation with direction-based initialization and time annotations using complex angular central Gaussian mixture models

Fig. 11

ΔSTOI and ΔPESQ scores for a moving speaker as a function of the total movement over the entire signal duration in terms of the azimuth angle φ. To obtain a clearer representation, a moving average was applied (averaging over all results that deviate by up to ±10∘ compared to the values indicated on the x-axis), and negative results were replaced by 0. For this setup, too, the initialization remains the most effective way to benefit from DOA information, rendering time annotations (STAs or DTAs) as well as manual permutation alignment unnecessary

Back to article page