Suppression of acoustic noise in speech using spectral subtraction

By J. Youngberg, T. Petersen, S. Boll, Elaine L. Cohen. 1979

How researchers see this paper

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

1. Xu Tan
2. Jinyu Li
3. Dong Yu
4. Wei-Ning Hsu
5. Deliang Wang
6. Li Deng
7. Yu Zhang
8. Shinji Watanabe

Citations in the corpus

• Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis (2018, methodology): “We preprocessed the target spectrogram using a simple spectral subtraction [4] denoising procedure, where the background noise spectrum of an utterance was estimated as the 10th percentile of the energy in each frequency band across the full signal.”
• Hierarchical Generative Modeling for Controllable Speech Synthesis (2018, methodology): “Specifically, we consider another stronger baseline model to compare on the SN set, which is trained on denoised data using spectral subtraction (Boll, 1979), denoted as “+ denoise.”
• A Hybrid DSP/Deep Learning Approach to Real-Time Full-Band Speech Enhancement (2017, background): “For example, the crude initial noise estimators and the spectral estimators based on spectral subtraction [1] have been replaced by more accurate noise estimators [2], [3] and spectral amplitude estimators [4].”
• Speech Enhancement in Multiple-Noise Conditions Using Deep Neural Networks (2016, methodology): “The simplest method to remove additive noise by subtracting an estimate of noise spectrum from noisy speech spectrum was proposed back in 1979 by Boll [3].”
• On the Importance of the Pearson Correlation Coefficient in Noise Reduction (2008, background): “In the single-channel scenario, the observed microphone signal can be modeled as a superposition of the clean speech and noise.”
• Supervised and Unsupervised Speech Enhancement Using Nonnegative Matrix Factorization (2013, methodology): “Unsupervised methods include a wide range of approaches such as spectral subtraction [1], Wiener and Kalman filterin g, e.”
• rVAD: An Unsupervised Segment-Based Robust Voice Activity Detection Method (2019, methodology): “Based on an estimate of additive-noise spectrum, a spectral subtraction (SS) method [53] is then used to subtract the estimate from the noisy speech spectrum.”
• Speech enhancement based on minimum mean-square error estimation and supergaussian priors (2005, background): “[1]–[4] assume that the DFT coefficients are statistically independent and that the real and imaginary part of the clean speech DFT coefficients can be modeled by a Gaussian density.”
• Weighted Speech Distortion Losses for Neural-Network-Based Real-Time Speech Enhancement (2020, background): “Although most classical SE methods have to accommodate their approaches [4, 5, 6, 7] for fulfilling causality, many DNN-based methods in the literature [3, 8, 9] do not enforce this constraint.”
• Histogram-based subband powerwarping and spectral averaging for robust speech recognition under matched and multistyle training (2012, background): “[1]), some mimic attributes of the human auditory system to enhance the properties of speech believed to be relevant to recognition (e.”
• Speech Enhancement for Noise-Robust Speech Synthesis Using Wasserstein GAN (2019, methodology): “Besides the standard signal processing approaches such as spectral subtraction and Wiener filter [1, 2], deep neural networks have been widely adopted to either directly reconstruct clean speech [3] or estimate masks [4] from the noisy signals.”
• Supervised Speech Separation Based on Deep Learning: An Overview (2017, unknown): “The simplest and most widely used enhancement method is spectral subtraction [13], in which the”
• Robust Features for Noisy Speech Recognition using MFCC Computation from Magnitude Spectrum of Higher Order Autocorrelation Coefficients (2010, methodology): “In the literature, various approaches have been proposed to improve the tolerance of an ASR system with respect to noise, such as Wiener filtering [2], spectral subtraction [3], RASTA [4], lin-log RASTA [5], parallel model compensation (PMC) [6], vector Taylor series approximation based model compen”
• State-Denoised Recurrent Neural Networks (2018, background): “A traditional approach to handling noise is to design preprocessing stages that estimate and filter noise from an input signal [Boll, 1979].”
• Multichannel Eigenspace Beamforming in a Reverberant Noisy Environment With Multiple Interfering Speech Signals (2009, methodology): “We note, however, that applying nonlinear postfilters such as spectral subtraction [40] or Optimally Modified Log Spectral Amplitude (OMLSA) [41] might be capable of further reducing the residual noise”
• An Overview of Noise-Robust Automatic Speech Recognition (2014, methodology): “a) Spectral Subtraction: The spectral subtraction [103] method assumes that noise and clean speech are uncorrelated and additive in the time domain.”