Speech Enhancement: Theory and Practice

By P. Loizou. 2007

2,560 citations recorded by the source corpus.

How researchers see this paper

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

In Signal Processing

Citations in the corpus

A Wavenet for Speech Denoising (2017, methodology): “We set as baselines: (i) the noisy signal, and (ii) a signal processing method based on Wiener filtering – widely used for speech-denoising [22, 23] or audio source-separation [24].”
SEGAN: Speech Enhancement Generative Adversarial Network (2017, background): “They only modify the spectrum magnitude, as it is often claimed that short-time phase is not important for speech enhancement [17].”
Full-Capacity Unitary Recurrent Neural Networks (2016, background): “PESQ is the ITU-T standard for telephone voice quality testing [21, 22], and is a popular perceptual quality metric for speech enhancement [23].”
SDR – Half-baked or Well Done? (2018, methodology): “While the merits of various objective measures such as PESQ [12], Loizou’s composite measure [13], PEMO-Q [14], PEASS [15], or STOI [16], could be debated and compared, we are concerned here with an issue with the way the widely relied upon BSS eval toolbox [17] has been used.”
ViSQOL: an objective speech quality model (2015, methodology): “The two best performing metrics from Hu and Loizou’s study, the log-likelihood ratio (LLR) and frequency domain segmental signal-to-noise ratio (fwSNRSeg) [29,30], are tested along with the specialised speech quality metrics, PESQ and POLQA, to illustrate their strengths and weaknesses.”
Speech Enhancement in Multiple-Noise Conditions Using Deep Neural Networks (2016, background): “it has been an actively researched topic and several methods have been proposed over the past several decades [1] [2].”
Subjective comparison and evaluation of speech enhancement algorithms (2007, methodology): “MATLAB implementations of all the algorithms tested are available in Loizou (2007).”
Bone-conducted speech enhancement using deep denoising autoencoder (2018, methodology): “Next, we use two standardized objective evaluation matrices, namely the perceptual evaluation of speech quality (PESQ) (Loizou, 2007) and short-time objective intelligibility (STOI) scores (Taal et al., 2011), respectively, to measure the speech quality and speech intelligibility of BCM, DDAE, and I”
Objective measures for predicting speech intelligibility in noisy conditions based on new band-importance functions. (2009, methodology): “It is also not known whether any of the numerous objective measures that have been proposed to predict speech quality Quackenbush et al., 1988; Loizou, 2007, Chap.”
Evaluation of Objective Quality Measures for Speech Enhancement (2008, methodology): “We kept only the subset of measures for which the correlation of the composite measure improved significantly from the correlation coefficient of the individual measures.”
NADiffuSE: Noise-aware Diffusion-based Model for Speech Enhancement (2023, background): “In the last decade, deep learning (DL) methods [1], [2] have become the mainstream for speech enhancement (SE), which can be divided into two categories: the discriminative and generative ones [3].”
Audio-Visual Speech Enhancement Using Multimodal Deep Convolutional Neural Networks (2017, methodology): “tion, the ambient noise for testing was a 60 mph car engine noise taken from the dataset used in [75], which was also different from those used in the training set.”
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.”
A Consolidated Perspective on Multimicrophone Speech Enhancement and Source Separation (2017, positive): “Implemented metrics PESQ [299] perceptual speech quality (PESQ) PEMO-Q [300] perceptual similarity metric (PSM) STOI [301] 27 short-time objective intelligibility (STOI) Loizou’s [11] implemented in Roomsimove 19 and Signal generator 20 .”
Data Augmentation for Robust Keyword Spotting under Playback Interference (2018, background): “…systems commonly fall into the categories of (a) signal or feature enhancement techniques that aim to remove the noise from the input features [9, 10] (b) model adaptation techniques that modify the model parameters to be more representative of the input speech [11] and (c) feature adaptation…”
A Review of Speaker Diarization: Recent Advances with Deep Learning (2021, methodology): “Also, the speech enhancement used in [95] applies this objective function in each layer based on a progressive manner.”
End-to-End Waveform Utterance Enhancement for Direct Evaluation Metrics Optimization by Fully Convolutional Neural Networks (2017, background): “1) As stated in the Introduction section, improving speech intelligibility is more challenging than enhancing quality [41], [42].”
On Training Targets for Supervised Speech Separation (2014, methodology): “However, they are generally less effective in low SNR and non-stationary noise conditions [21].”
Deep Learning for Environmentally Robust Speech Recognition (2017, methodology): “ng speech intelligibility but this appears to depend on the noise conditions [52]. In the literature, many unsupervised signal processing methods have been developed for removing non-stationary noise [4]. The advantage of these methods over supervised approaches is that they do not require training ”