Super-human multi-talker speech recognition: A graphical modeling approach

By John R. Hershey, Steven J. Rennie, Peder A. Olsen, Trausti T. Kristjansson. 2010

166 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

How researchers see this paper

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

In Artificial Intelligence

In Signal Processing

Citations in the corpus

End-to-End Speech Separation with Unfolded Iterative Phase Reconstruction (2018, methodology): “The inventions of deep clustering [1, 2, 3], deep attractor networks [4, 5] and permutation free training [1, 2, 6, 7] have dramatically improved the performance of single-channel speaker-independent multispeaker speech separation, demonstrating overwhelming advantages over previous methods includin”
Neural Target Speech Extraction: An overview (2023, methodology): “An F-HMM is a model of speech mixtures, where the speech of each speaker in the mixture is explicitly modeled using a separate hidden Markov model (HMM).”
Cracking the cocktail party problem by multi-beam deep attractor network (2017, unknown): “…systems, the separation process entirely relies on the spectral properties of speech, such as pitch continuity, harmonic structures, common onsets etc., and this can be achieved by using statistical models [10], rule-based models [3, 11], or decomposition-based models [12, 13, 14, 15].”
On Training Targets for Supervised Speech Separation (2014, methodology): “For example, techniques in [24], [13] build probabilistic interaction models between different sources based on learned priors, and show significant performance gain in low SNR conditions.”
Alternative Objective Functions for Deep Clustering (2018, methodology): “…neural networks, all of these approaches have demonstrated overwhelming advantages over previous approaches including graphical modeling approaches [6], spectral clustering approaches [7], and CASA methods [8] In this study, we introduce several alternative deep clustering objectives that aim to…”
RETRIEVE: Coreset Selection for Efficient and Robust Semi-Supervised Learning (2021, background): “1 Introduction Deep learning algorithms have had great success over the past few years, often achieving human or superhuman performance in various tasks like computer vision [9], speech recognition [17], natural language processing [5], and video games [38].”
Multi-Channel Deep Clustering: Discriminative Spectral and Spatial Embeddings for Speaker-Independent Speech Separation (2018, methodology): “…was the ﬁrst modern approach to perform speaker-independent separation, and demonstrated superior performance over previous attempts at speech separation, including graphical modeling approaches [3], spectral clustering approaches [4], and computational auditory scene analysis based methods [5].”
The PASCAL CHiME speech separation and recognition challenge (2013, background): “The challenge was remarkable in that the best-performing system was able to produce super-human performance (Hershey et al., 2010).”
Single Channel Target Speaker Extraction and Recognition with Speaker Beam (2018, background): “Solving such mismatch between a speech enhancement front-end and an ASR back-end has been addressed by jointly training both modules [15, 21, 22].”
Deep neural networks for single channel source separation (2013, methodology): “Fixing this limitation requires complicated computations as in [7, 8, 9, 10, 11, 12].”
Deep clustering: Discriminative embeddings for segmentation and separation (2015, methodology): “In another direction, [8] extended [5] to handle more than two speakers.”
Single-Channel Multi-Speaker Separation Using Deep Clustering (2016, methodology): “In constrained tasks, super-human speech separation was ﬁrst achieved using factorial HMMs [8, 9, 10, 11] and was extended to speaker-independent separation [12].”