Noise spectrum estimation in adverse environments: improved minima controlled recursive averaging

By Israel Cohen. IEEE Transactions on Speech and Audio Processing, 2003

986 citations recorded by the source corpus.

Subfields: Cognitive Neuroscience, Computational Mechanics, Signal Processing

View Paper: https://www.semanticscholar.org/paper/9ac25bd4927135e515967de8aafb5e8b3740c391

How researchers see this paper

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

In Cognitive Neuroscience

In Computational Mechanics

In Signal Processing

Citations in the corpus

DeepMMSE: A Deep Learning Approach to MMSE-Based Noise Power Spectral Density Estimation (2020, methodology): “Another group of noise PSD tracking methods are time-recursive averaging algorithms, including the minima controlled recursive averaging (MCRA) method [15], the improved MCRA method (IMCRA) [16], and MCRA-2 [17].”
A Regression Approach to Speech Enhancement Based on Deep Neural Networks (2015, methodology): “An improved version of OM-LSA [8], [9], denoted as LogMMSE, was used for performance comparison”
Improved cepstra minimum-mean-square-error noise reduction algorithm for robust speech recognition (2017, methodology): “In [29], an improved MCRA (IMCRA) algorithm is shown to outperform MCRA.”
Subjective comparison and evaluation of speech enhancement algorithms (2007, methodology): “logMMSE algorithm when a noise-estimation algorithm ( Cohen, 2003 ) was used to update the noise spectrum.”
Investigating RNN-based speech enhancement methods for noise-robust Text-to-Speech (2016, positive): “As a conventional speech enhancement method we choose the statistical model-based method described in [31] that uses the optimally-modified log-spectral amplitude speech estimator (OMLSA) and an improved version of the minima controlled recursive averaging noise estimator as proposed in [32].”
An Experimental Study on Speech Enhancement Based on Deep Neural Networks (2014, methodology): “To evaluate the performance of DNN-based speech enhancement, an improved version of the optimallymodified log-spectral amplitude (OMLSA) [2], [15], [16], denoted as log-MMSE (L-MMSE) method, was used for performance comparison.”
Speech Enhancement for a Noise-Robust Text-to-Speech Synthesis System Using Deep Recurrent Neural Networks (2016, methodology): “As a conventional speech enhancement method we choose the method described in [30] that uses the optimally-modified log-spectral amplitude speech estimator (OMLSA) and an improved version of the minima controlled recursive averaging noise estimator as proposed in [31].”
Speech Enhancement of Noisy and Reverberant Speech for Text-to-Speech (2018, methodology): “This method uses the optimallymodified log-spectral amplitude speech estimator (OMLSA) and the minima controlled recursive averaging noise estimator as proposed in [33].”
Multi-objective learning and mask-based post-processing for deep neural network based speech enhancement (2017, methodology): “An improved version of OM-LSA [5], denoted as LogMMSE , was used for performance comparison with our DNN approach.”
Deep Learning for Environmentally Robust Speech Recognition (2017, methodology): “ is strongly affected by the noise modelling approach, more sophisticated noise estimation techniques have also been developed, for example, tracking spectral minima also during speech activity [53], [54]. 4 Whilst an instantaneous representation of the noise is obtained, noise reduction methods typ”
Unbiased MMSE-Based Noise Power Estimation With Low Complexity and Low Tracking Delay (2012, methodology): “In contrast to [16], these fixed priors are independent of the observation.”
Speech enhancement based on deep denoising autoencoder (2013, methodology): “In our comparison, we took the MMSE plus improved minimum controlled recursive averaging (IMCRA) noise tracking algorithm [13].”
A noise-estimation algorithm for highly non-stationary environments (2006, methodology): “In (Cohen, 2003), non-stationary noise was modeled by increasing the level of WGN by 2 dB/s.”
Combining missing-feature theory, speech enhancement, and speaker-dependent/-independent modeling for speech separation (2006, neutral): “Each test TMR condition contains 600 test utterances, of which, one third are masked by the same talker, one third are masked by talkers of the same gender, and the remaining are masked by talkers of different genders.”