Fig. 3From: Introducing phonetic information to speaker embedding for speaker verificationHybrid multi-task learning. The phonetic-discriminant network shares layers with the frame-level part of the x-vector architectureBack to article page