The Second DIHARD Challenge: System Description for USC-SAIL Team

Abstract

In this paper, we describe components that form a part of USC-SAIL team's submissions to Track 1 and Track 2 of the second DIHARD speaker diarization challenge. We describe each module in our speaker diarization pipeline and explain the rationale behind our choice of algorithms for each module, while comparing the Diarization Error Rate (DER) against different module combinations. We propose a clustering scheme based on spectral clustering that yields competitive performance. Moreover, we introduce an overlap detection scheme and a re-segmentation system for speaker diarization and investigate their performances using controlled and in-the-wild conditions. In addition, we describe the additional components that will be integrated to our speaker diarization system. To pursue the best performance, we compare our system with the stateof-the-art methods that are presented in the previous challenge and literature. We include preliminary results of our speaker diarization system on the evaluation data from the second DI-HARD challenge.

Speaker Diarization Using Latent Space Clustering in Generative Adversarial Network

“…In terms of clustering, most of the existing algorithms that have been used in speaker diarization are unsupervised. Among them, agglomerative hierarchical clustering (AHC) [3] and spectral clustering (SC) [16] using pairwise embedding similarity measurement techniques like cosine distance [6], [9], PLDA [17] and using an LSTM [10] are the most popular. Similarly, other unsupervised clustering methods such as Gaussian mixture model [4], [13], mean-shift [5], k-means [18], and links [19] have also been adopted for speaker diarization.…”