Figure 5 | Scientific Reports

Figure 5

From: Recurrence plot embeddings as short segment nonlinear features for multimodal speaker identification using air, bone and throat microphones

Figure 5

The speaker recognition results using the CV dataset. (a) compares the accuracy of the proposed RP embedding as a stand-alone nonlinear feature against other nonlinear features mentioned in literature for the vowel-specific speaker recognition system, and (b) shows the good generalization performance of all transfer learning models for the speaker recognition study using the overall (all CV utterances, irrespective of vowels, of each speaker pooled together to form his/her data, as opposed to that mentioned in (a)) CV dataset.

Back to article page