speaker_recognition_papers.md

Speaker Recognition Papers (Deep Learning Era)

• Dataset
  • VoxCeleb
    • VoxCeleb: a large-scale speaker identification dataset
    • VoxCeleb2: Deep Speaker Recognition
    • Voxceleb: Large-scale speaker verification in the wild
  • CnCeleb
    • CN-CELEB: A CHALLENGING CHINESE SPEAKER RECOGNITION DATASET
    • CN-Celeb: multi-genre speaker recognition
• Architecture Design
  • X-VECTORS: ROBUST DNN EMBEDDINGS FOR SPEAKER RECOGNITION (x-vector)
  • BUT System Description to VoxCeleb Speaker Recognition Challenge 2019 (r-vector)
  • RawNet: Advanced end-to-end deep neural network using raw waveforms for text-independent speaker verification (RawNet)
  • SPEAKER RECOGNITION FROM RAW WAVEFORM WITH SINCNET (SincNet)
  • ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification (ECAPA-TDNN)
• Optimization Objective
  • Classification Based Loss
    • Exploring the encoding layer and loss function in end-to-end speaker and language recognition system
    • Angular Softmax for Short-Duration Text-independent Speaker Verification
    • ENSEMBLE ADDITIVE MARGIN SOFTMAX FOR SPEAKER VERIFICATION
    • Margin Matters: Towards More Discriminative Deep Neural Network Embeddings for Speaker Recognition
    • Large Margin Softmax Loss for Speaker Verification
  • End-to-End Loss
    • End-to-End Text-Dependent Speaker Verification
    • End-to-End Text-Independent Speaker Verification with Triplet Loss on Short Utterances
    • GENERALIZED END-TO-END LOSS FOR SPEAKER VERIFICATION
• Pooling Method
  • Attentive Statistics Pooling for Deep Speaker Embedding
  • Multi-Resolution Multi-Head Attention in Deep Speaker Embedding
  • Utterance-level Aggregation for Speaker Recognition in the Wild
  • A novel learnable dictionary encoding layer for end-to-end language identification
• Self-supervised Learning
  • Augmentation adversarial training for self-supervised speaker recognition
  • SELF-SUPERVISED TEXT-INDEPENDENT SPEAKER VERIFICATION USING PROTOTYPICAL MOMENTUM CONTRASTIVE LEARNING
  • SELF-SUPERVISED SPEAKER RECOGNITION WITH LOSS-GATED LEARNING