• Kaldi recipe for speaker recognition using xvectors (GitHub) [code]:
            The recipe shows how to train a DNN to compute speaker embeddings (xvectors). The system is evaluated on the NIST-SRE16 setup.
  • Kaldi recipe for DNN-based speaker recognition (GitHub) [code]:
             The recipe shows how to replace an unsupervised GMM-UBM with a DNN that was trained on transcribed data to classify senones. The system is evaluated on the NIST-SRE10 setup.
This package contains a Matlab implementation of Gaussian-PLDA with length normalization. The compressed tar ball contains all the necessary i-vectors (i.e., development, model and test) to produce results for the NIST-SRE10 extended evaluation in the 1conv-1conv setup.