基于x-vector嵌入与BLSOM模型的声纹聚类系统

Voiceprint Clustering System Based on x-vector Embedding and BLSOM Model

介绍一种基于无监督神经网络的声纹识别方法。提取语音Filter Banks特征并将其输入神经网络特征提取器x-vec tor提取紧凑的较低维度说话人表征特征向量,并采用批处理自组织神经网络BLSOM进行自适应聚类训练。实验结果表明,该方法具有较好的声纹聚类识别效果。

Introduces a voiceprint recognition method based on unsupervised neural network.Speech Filter Banks features are extracted and input to the deep learning feature extractor x-vector to extract compact lower-dimension speaker characterization feature vectors,and batch self-or ganizing neural network BLSOM is used for adaptive cluster training.The experimental results show that the method has better recognition effect of voiceprint clustering.