Alternative Objective Functions for Deep Clustering

By Zhong-Qiu Wang, Jonathan Le Roux, John R. Hershey. 2018

191 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Signal Processing

Citations in the corpus

Single Channel Voice Separation for Unknown Number of Speakers Under Reverberant and Noisy Settings (2020, methodology): “In this work, we focus on fully supervised voice separation using a single microphone, which has seen a great leap in performance following the recent success of deep learning models considering both frequency domain [1, 2, 7, 8, 9, 10], and time-domain [11, 12, 13, 14, 15, 16] modeling.”
Voice Separation with an Unknown Number of Multiple Speakers (2020, methodology): “Wang et al. (2018c) propose a novel method to separate speech signals in the time-frequency domain.”
SDR – Half-baked or Well Done? (2018, methodology): “5 Chimera++ Networks [11] 11.”
Multi-Channel Deep Clustering: Discriminative Spectral and Spatial Embeddings for Speaker-Independent Speech Separation (2018, background): “Our recent study [20] found that further improvements could be obtained using an alternative cost function based on whitening the embeddings in a k-means objective:”
FurcaNeXt: End-to-end monaural speech separation with dynamic gated dilated temporal convolutional networks (2019, unknown): “8 Chimera++ [31] 12.”
WHAMR!: Noisy and Reverberant Single-Channel Speech Separation (2019, methodology): “Great advancements have been made in recent years on solving the speech separation problem through deep learning-based techniques [1–6].”
The Phasebook: Building Complex Masks via Discrete Representations for Source Separation (2019, methodology): “Recent works applied phase reconstruction as a post-processing [11], then as part of the optimization pipeline [14].”
Filterbank Design for End-to-end Speech Separation (2019, unknown): “The results are reported in Table 3 along with the chimera++ [6] results in [14].”
Acoustic Modeling for Distant Multi-talker Speech Recognition with Single- and Multi-channel Branches (2019, methodology): “We internally evaluated deep clustering [19], permutation invariant training [20], chimera++ networks [21], and a speakeraware neural beamformer [22, 23], but none of them successfully improved the accuracy for the CHiME-5 dataset.”
Towards Neural Diarization for Unlimited Numbers of Speakers Using Global and Local Attractors (2021, background): “Some neural-network-based speech separation methods [19, 39, 40, 41, 42] limit the number of outputs by their network architecture, and thus, there is no way to deal with the mismatch.”
Universal Sound Separation (2019, methodology): “Recent advances have been made in solving monaural speech enhancement and speech separation problems in increasingly difﬁcult scenarios, thanks to deep learning methods [1–11].”
Dual-Path Transformer Network: Direct Context-Aware Modeling for End-to-End Monaural Speech Separation (2020, unknown): “SI-SNR SDR Model Size DPCL++ [27] 10.8 - 13.6M uPIT-BLSTM-ST [12] - 10.0 92.7M Deep Attractor [10] 10.5 - - ADANet [28] 10.4 10.8 9.1M Grid LSTM PIT [29] - 10.2 - ConvLSTM-GAT [30] - 11.0 - Chimera++ [31] 11.5 12.0 - WA-MISI-5 [32] 12.6 13.1 32.9M BLSTM-TasNet [14] 13.2 13.6 - Conv-TasNet-gLN [15] 1”
Deep Learning Based Phase Reconstruction for Speaker Separation: A Trigonometric Perspective (2018, methodology): “One straightforward approach for phase reconstruction is to use estimated magnitudes to drive an iterative phase reconstruction algorithm [15], [16], [5], [6].”
WHAM!: Extending Speech Separation to Noisy Environments (2019, methodology): “For speech separation, mask estimation can be further improved by incorporating a deep clustering regularization term into the loss function as in the chimera++ architecture [11], i.e., where the weight α is typically set to a high value, e.g., 0.975.”
Recursive speech separation for unknown number of speakers (2019, background): “Recent advances of deep learning based methods including deep clustering (DPCL) [11–14], permutation invariant training (PIT) [15–17], deep attractor network (DANet) [18, 19] dramatically improved the accuracy of separation.”
Bootstrapping Single-channel Source Separation via Unsupervised Spatial Clustering on Stereo Mixtures (2018, methodology): “Because this pseudo ground truth may not be reliable, we use a version of the original DC objective [6] that was modified to include a weight wi for each T-F bin i = (t, f) [26]:”