Convolutive Speech Bases and Their Application to Supervised Speech Separation

By Paris Smaragdis. IEEE Transactions on Audio, Speech, and Language Processing, 2007

425 citations recorded by the source corpus.

Subfields: Computational Mechanics, Signal Processing

View Paper: https://www.semanticscholar.org/paper/150bbc9c982c9f99e956444473cd7b99295cebb9

How researchers see this paper

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

In Computational Mechanics

1. Deliang Wang
2. Jinyu Li
3. Zhuo Chen
4. Dong Yu
5. John R. Hershey
6. Jonathan Le Roux
7. Shinji Watanabe
8. Yi Luo

In Signal Processing

1. Jinyu Li
2. Dong Yu
3. Shinji Watanabe
4. Zhuo Chen
5. Deliang Wang
6. John R. Hershey
7. Yi Luo
8. Nima Mesgarani

Citations in the corpus

• Single-Channel Multi-Speaker Separation Using Deep Clustering (2016, methodology): “Variants of non-negative matrix factorization [13, 14] and Bayesian non-parametric models [15, 16] have also been used.”
• Exemplar-Based Sparse Representations for Noise Robust Automatic Speech Recognition (2011, methodology): “Algorithms that can estimate the activations in such a model can be based on the convolutive NMF approaches described in [45] and [46].”
• Nonnegative Matrix Factorization: A Comprehensive Review (2013, methodology): “Both GKLD [111], [112] and SED [22] objec-”
• An Overview of Noise-Robust Automatic Speech Recognition (2014, background): “Convolutive extension of NMF is proposed to handle potential dependencies across successive input columns [171], [173].”
• FurcaNeXt: End-to-end monaural speech separation with dynamic gated dilated temporal convolutional networks (2019, methodology): “Many traditional methods, such as computational auditory scene analysis (CASA) [30, 20, 10], Non-negative matrix factorization (NMF) [23, 13], and probabilistic models [29], do not solve these two difficulties well.”
• Speaker-Aware Neural Network Based Beamformer for Speaker Extraction in Speech Mixtures (2017, methodology): “This problem has been investigated using both single and multi-channel source separation techniques, including NMF [1], ICA [2] or clustering of spatial cues [3, 4, 5].”
• Direction of Arrival Based Spatial Covariance Model for Blind Sound Source Separation (2014, neutral): “More recently, methods based on finding spectrally redundant parts by non-negative matrix factorization (NMF) [12]–[14] have been proposed for separation of sound sources both with single [15], [16] and multichannel mixtures [17]–[20].”
• Permutation invariant training of deep models for speaker-independent multi-talker speech separation (2016, methodology): “This mask is then used to reconstruct the signaL Non-negative matrix factorization (NMF) [12, 13, 14] is another popular technique which aims to learn a set of non-negative bases that can be used to estimate mixing factors during evaluation.”
• Improving Robustness of Deep Neural Network Acoustic Models via Speech Separation and Joint Adaptive Training (2015, background): “Systems like convolutional NMF [36] and structure-preserving models [43] only account for short term sequence structure.”
• Speaker-Independent Speech Separation With Deep Attractor Network (2017, background): “as computational auditory scene analysis (CASA) [11], [12] and spectral clustering [13] fall into this category [14]. Factorization models, such as non-negative matrix factorization (NMF) [15], [16], [17], formulate the separation problem as a matrix factorization problem in which the timefrequency ”
• CBLDNN-Based Speaker-Independent Speech Separation Via Generative Adversarial Training (2018, methodology): “Non-negative matrix factorization (NMF) based methods [3, 4, 5] decompose signal into sets of bases and weight matrices, which are used to estimate mixing factors during evaluation.”
• Single Channel Speech Separation with Constrained Utterance Level Permutation Invariant Training Using Grid LSTM (2018, methodology): “Another non-negative matrix factorization (NMF) [4, 5, 6] techniques reconstruct each source using non-negative dictionaries and activations.”