LEAP Diarization System for the Second DIHARD Challenge

Singh, Prachi; Vardhan, Harsha; Ganapathy, Sriram; Kanagasundaram, Ahilan

doi:10.21437/interspeech.2019-2716

Cited by 11 publications

(13 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The diarization baseline is based on LEAP Lab's submission to DIHARD II [27]. The system performs diarization by dividing each recording into short overlapping segments, extracting x-vectors [28,29], scoring with probabilistic linear discriminant analysis (PLDA) [30], and clustering using agglomerative hierarchical clustering (AHC) [31].…”

Section: Diarizationmentioning

confidence: 99%

“…The system performs diarization by dividing each recording into short overlapping segments, extracting x-vectors [28,29], scoring with probabilistic linear discriminant analysis (PLDA) [30], and clustering using agglomerative hierarchical clustering (AHC) [31]. The AHC ouput is then refined using Variational Bayes Hidden Markov Model (VB-HMM) [32,33] with posterior scaling [27]. The trained models and recipes for both tracks are distributed through GitHub 5 .…”

Section: Diarizationmentioning

confidence: 99%

“…Both the UBM-GMM and V were trained using the same data as for the x-vector extractor. Following [27], posterior scaling was applied to discourage frequent speaker transitions by the VB-HMM. This scaling was accomplished by boosting the zeroth order, but not first or second order, statistics prior to VB-HMM likelihood computation.…”

Section: Diarizationmentioning

confidence: 99%

See 2 more Smart Citations

The Third DIHARD Diarization Challenge

et al. 2021

Self Cite

View full text Add to dashboard Cite

This paper introduces the third DIHARD challenge, the third in a series of speaker diarization challenges intended to improve the robustness of diarization systems to variation in recording equipment, noise conditions, and conversational domain. Speaker diarization is evaluated under two segmentation conditions (diarization from a reference speech segmentation vs. diarization from scratch) and 11 diverse domains. The domains span a range of recording conditions and interaction types, including read audiobooks, meeting speech, clinical interviews, web videos, and, for the first time, conversational telephone speech. We describe the task and metrics, challenge design, datasets, and baseline systems for speech speech activity detection and diarization.

show abstract

Section: Diarizationmentioning

confidence: 99%

Section: Diarizationmentioning

confidence: 99%

See 1 more Smart Citation

The Third DIHARD Diarization Challenge

et al. 2021

Self Cite

View full text Add to dashboard Cite

show abstract

“…The proposed approach also shows advancements over other published results on these datasets. Furthermore, the proposed approach can be used as an initialization for frame-level refinement based on variational Bayes (VB) hidden Markov model (HMM) [26], [27].…”

Section: Ti ( Ti )mentioning

confidence: 99%

“…In the end-to-end learning, the input features are fed to a model where the loss is either permutation-invariant cross entropy [39], [40] or clustering based [41]. Further to refine the boundaries of segmentation output in speaker diarization, a second re-segmentation step involving frame-level (20-30ms) modeling [26], [27] can be performed. This model the variational-Bayes HMM model which is typically initialized using a segment level diarization output [42].…”

Section: Related Workmentioning

confidence: 99%

Deep Self-Supervised Hierarchical Clustering for Speaker Diarization

Singh¹,

Ganapathy²

2020

Interspeech 2020

Self Cite

View full text Add to dashboard Cite

Automatic speaker diarization techniques typically involve a two-stage processing approach where audio segments of fixed duration are converted to vector representations in the first stage. This is followed by an unsupervised clustering of the representations in the second stage. In most of the prior approaches, these two stages are performed in an isolated manner with independent optimization steps. In this paper, we propose a representation learning and clustering algorithm that can be iteratively performed for improved speaker diarization. The representation learning is based on principles of selfsupervised learning while the clustering algorithm is a graph structural method based on path integral clustering (PIC). The representation learning step uses the cluster targets from PIC and the clustering step is performed on embeddings learned from the self-supervised deep model. This iterative approach is referred to as self-supervised clustering (SSC). The diarization experiments are performed on CALLHOME and AMI meeting datasets. In these experiments, we show that the SSC algorithm improves significantly over the baseline system (relative improvements of 13% and 59% on CALLHOME and AMI datasets respectively in terms of diarization error rate (DER)). In addition, the DER results reported in this work improve over several other recent approaches for speaker diarization.

show abstract