Fig. 3From: Single microphone speech separation by diffusion-based HMM estimationThe DFHMM state inference scheme. The states are inferred by applying two FHMMs. Each FHMM infers the states of the speaker whose training set was used to compute its embeddingBack to article page