MetricGAN+/-: Increasing Robustness of Noise Reduction on Unseen Data

Close, George; Hain, Thomas; Goetze, Stefan

doi:10.23919/eusipco55093.2022.9909682

Cited by 7 publications

(6 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The overall architecture of the proposed system is based on the conformer-based metric GAN (CMGAN) framework proposed in [13], but with two extensions based on [14] and [15]. The first extension is to train the discriminator D on a historical set of past generator outputs every epoch.…”

Section: Speech Enhancement Systemmentioning

confidence: 99%

Multi-CMGAN+/+: Leveraging Multi-Objective Speech Quality Metric Prediction for Speech Enhancement

Close,

Ravenscroft,

Hain

et al. 2024

ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

Neural network based approaches to speech enhancement have shown to be particularly powerful, being able to leverage a data-driven approach to result in a significant performance gain versus other approaches. Such approaches are reliant on artificially created labelled training data such that the neural model can be trained using intrusive loss functions which compare the output of the model with clean reference speech. Performance of such systems when enhancing real-world audio often suffers relative to their performance on simulated test data. In this work, a non-intrusive multi-metric prediction approach is introduced, wherein a model trained on artificial labelled data using inference of an adversarially trained metric prediction neural network. The proposed approach shows improved performance versus state-of-the-art systems on the recent CHiME-7 challenge unsupervised domain adaptation speech enhancement (UDASE) task evaluation sets.

show abstract

Section: Speech Enhancement Systemmentioning

confidence: 99%

Multi-CMGAN+/+: Leveraging Multi-Objective Speech Quality Metric Prediction for Speech Enhancement

Close,

Ravenscroft,

Hain

et al. 2024

ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

show abstract

“…From this, the enhanced time domain speech ŝ[n] is then created via the overlap-add resynthesis method, using the original noisy phase of x[n]. This structure is selected because, despite being relatively simple and with a small parameter count, it is able to achieve state of the art performance in perceptually motivated speech enhancement [2,5,10].…”

Section: Enhancement Model Structurementioning

confidence: 99%

“…Speech enhancement remains an active area of speech research due to its applications in numerous downstream tasks [1]. Deep learning models have led to state-of-the-art results on numerous benchmarks for speech enhancement and related tasks [2][3][4][5]. A key area of research for these kinds of models has been the loss functions used [6].…”

Section: Introductionmentioning

confidence: 99%

Perceive and Predict: Self-Supervised Speech Representation Based Loss Functions for Speech Enhancement

Ravenscroft

Hain

et al. 2023

ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Self Cite

View full text Add to dashboard Cite

Recent work in the domain of speech enhancement has explored the use of self-supervised speech representations to aid in the training of neural speech enhancement models. However, much of this work focuses on using the deepest or final outputs of self supervised speech representation models, rather than the earlier feature encodings. The use of self supervised representations in such a way is often not fully motivated. In this work it is shown that the distance between the feature encodings of clean and noisy speech correlate strongly with psychoacoustically motivated measures of speech quality and intelligibility, as well as with human Mean Opinion Score (MOS) ratings. Experiments using this distance as a loss function are performed and improved performance over the use of STFT spectrogram distance based loss as well as other common loss functions from speech enhancement literature is demonstrated using objective measures such as perceptual evaluation of speech quality (PESQ) and short-time objective intelligibility (STOI).

show abstract

“…As such, the development of methods to increase speech intelligibility (SI) in assistive listening devices to alleviate this is of paramount importance [3]. While there have been large improvements in speech enhancement technology thanks to neural network-based approaches [4][5][6] these can often be challenging to implement in small form THIS WORK WAS SUPPORTED BY factor hearing aid (HA) hardware. Furthermore, given that the exact severity and nature of hearing loss differs greatly between individuals, a 'one size fits all' approach is not viable.…”

Section: Introductionmentioning

confidence: 99%

“…shows the Spearman and Pearson correlations of the MSE distances (4)-(6) with the correctness values i for the CPC1 training set. Absolute correlations are low, but this…”

mentioning

confidence: 99%

Non-intrusive Speech Intelligibility Metric Prediction for Hearing Impaired Individuals

Close¹,

Hollands²,

Goetze³

et al. 2022

Interspeech 2022

View full text Add to dashboard Cite

This paper proposes neural models to predict Speech Intelligibility (SI),both by prediction of established SI metrics and of human speech recognition (HSR) on the 1st Clarity Prediction Challenge. Both intrusive and non-intrusive predictors for intrusive SI metrics are trained, then fine-tuned on the HSR ground truth. Results are reported on a number of SI metrics, and the model choice for the Clarity challenge submission is explained. Additionally, the relationship between the SI scores in the data and commonly used signal processing metrics which approximate SI are analysed, and some issues emerging from this relationship discussed. It is found that intrusive neural predictors of SI metrics when fine-tuned on the true HSR scores outperform the non neural challenge baseline.

show abstract

MetricGAN+/-: Increasing Robustness of Noise Reduction on Unseen Data

Abstract: This is a repository copy of MetricGAN+/-: increasing robustness of noise reduction on unseen data.

Cited by 7 publications

References 20 publications

Multi-CMGAN+/+: Leveraging Multi-Objective Speech Quality Metric Prediction for Speech Enhancement

Multi-CMGAN+/+: Leveraging Multi-Objective Speech Quality Metric Prediction for Speech Enhancement

Perceive and Predict: Self-Supervised Speech Representation Based Loss Functions for Speech Enhancement

Non-intrusive Speech Intelligibility Metric Prediction for Hearing Impaired Individuals

Contact Info

Product

Resources

About