MelID is a ConvLSTM-based model for speaker ID/verification tasks, trained with Triplet Loss for embeddings generation.