Acoustical and environmental robustness in automatic speech recognition

By A. Acero. 1991

How researchers see this paper

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

1. Li Deng
2. Dong Yu
3. Jinyu Li
4. Yoshua Bengio
5. Song Han
6. Hongxu Yin
7. Xiaoliang Dai
8. Ji Lin

Citations in the corpus

• Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training (2017, methodology): “We validate our approach on three types of machine learning tasks: image classification on Cifar10 and ImageNet, language modeling on Penn Treebank dataset, and speech recognition on AN4 and Librispeech corpus.”
• Large-vocabulary speech recognition under adverse acoustic environments (2000, methodology): “The SPLICE algorithm outlined above is a modification and extension of the FCDCN algorithm described in [ 3 ].”
• High-performance robust speech recognition using stereo training data (2001, methodology): “Similar comparisons were made between SPLICE and FCDCN (described in [ 2 ] where acoustic partitioning was performed in the space of clean speech).”
• An Overview of Noise-Robust Automatic Speech Recognition (2014, methodology): “Fixed codeword-dependent cepstral normalization (FCDCN) [5] is a re fi ned version of SDCN with the compensation vector as , which depends on both SNR and VQ cluster location.”
• Improved cepstra minimum-mean-square-error noise reduction algorithm for robust speech recognition (2017, background): “INTRODUCTION Environment robustness in automatic speech recognition (ASR) remains a difficult problem despite many years of research [1][2][3][4][5][6][7][8].”
• Efficient Cepstral Normalization for Robust Speech Recognition (1993, methodology): “SDCN. The simplest compensation algorithm, SNR-Dependent Cepstral Normalization (SDCN) [2, 4 ], applies an additive correction in the cepstral domain that depends exclusively on the instantaneous SNR of the signal.”
• HMM adaptation using vector taylor series for noisy speech recognition (2000, methodology): “There are three main variants of the PMC method [1] depending on how the means and covariance matrices are computed.”
• Extended VTS for noise-robust speech recognition (2009, background): “In the time domain this has the form [9]”
• Model-Based Approaches to Handling Uncertainty (2011, background): “The standard, simplified model of the impact background additive noise, nt , and convolutional distortion, ht on the clean xt , is [1]”
• RyanSpeech: A Corpus for Conversational Text-to-Speech Synthesis (2021, background): “These corpora are well-suited to automatic speech recognition systems that require variations in input data [4].”
• Sources of degradation of speech recognition in the telephone network (1994, methodology): “The RASTA [6] and CDCN [1] procedures are two examples of algorithms that have been developed to compensate for the effects of linear filtering and additive noise in office environments.”
• A Linear Speedup Analysis of Distributed Deep Learning with Sparse and Quantized Communication (2018, methodology): “For speech recognition, we train a 5-layer LSTM of 800 hidden units per layer on AN4 dataset [2].”
• Nonlinear Compensation Using the Gauss–Newton Method for Noise-Robust Speech Recognition (2012, background): “Finally, conclusions are given in Section VIII.”
• Deep Learning for Environmentally Robust Speech Recognition (2017, unknown): “In Section VI, we then review promising multi-channel algorithms, before drawing our conclusions in Section VII.”
• Incremental Learning Using a Grow-and-Prune Paradigm With Efficient Neural Networks (2019, methodology): “Dataset: The speech recognition dataset in our experiment is the AN4 dataset [40], which has 948 and 130 utterances for training and validation, respectively.”
• Extended VTS for Noise-Robust Speech Recognition (2011, background): “In the time domain this has the form [9] (1) where denotes convolution.”