Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing
Fig. 14 | EURASIP Journal on Audio, Speech, and Music Processing

Fig. 14

From: Robustness of ad hoc microphone clustering using speaker embeddings: evaluation under realistic and challenging scenarios

Fig. 14

Incorporation of known speaker embedding for targeted clustering. In this scenario, the interfering source is active throughout the experiment. However, the target (known) source is only active for a short period in the middle of the segment. This is schematically indicated above, where the blue bar indicates the time-period where the known speaker is active, while red indicates the interferer activity. We initialise the first cluster with the known speaker embedding. Thereby, the FCM algorithm generates an empty cluster in parts a and c and only allocates microphones to the target cluster when the known source is active. In the figure, a microphone is part of the cluster if its colour is dark purple, while the light blue colour indicated that it is not part of the cluster

Back to article page