Skip to main content
Fig. 1 | EURASIP Journal on Advances in Signal Processing

Fig. 1

From: Features for voice activity detection: a comparative analysis

Fig. 1

Example of evaluation of the detection (gray areas) of a VAD algorithm by determining errors (striped areas) compared to the reference. a For ROC curves, VAD results are averaged over reference speech and noise intervals to determine probability of detection (P d) and false alarm (P fa). b The measure by Freeman et al. [3] considers four intervals, bounded by reference and detection end points, to determine the FEC, MSC, OVER, and NDS. c The fine-grained measure increases the temporal selectivity by a frame-wise evaluation around reference speech on- and offsets [4]. The dynamic behavior is captured by averaging only over utterances but not over time

Back to article page