Combining Mask Estimates for Single Channel Audio Source Separation Using Deep Neural Networks

Grais, Emad M.; Roma, Gerard; Simpson, Andrew; Plumbley, Mark D.

doi:10.21437/interspeech.2016-216

Cited by 31 publications

(28 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Nowadays, deep learning solutions seem to be the new El Dorado for audio processing. Still, most studies deal with single-channel denoising / enhancement / separation algorithms [102,22,51,55,139,86]. More recently, multichannel processing solutions that employ deep learning [104,23], as well as robust ASR systems [60], have been proposed.…”

Section: Discussionmentioning

confidence: 99%

Audio source separation into the wild

Girin

Gannot

2019

Multimodal Behavior Analysis in the Wild

View full text Add to dashboard Cite

This review chapter is dedicated to multichannel audio source separation in real-life environment. We explore some of the major achievements in the field and discuss some of the remaining challenges. We will explore several important practical scenarios, e.g. moving sources and/or microphones, varying number of sources and sensors, high reverberation levels, spatially diffuse sources, and synchronization problems. Several applications such as smart assistants, cellular phones, hearing aids and robots, will be discussed. Our perspectives on the future of the field will be given as concluding remarks of this chapter.

show abstract

Section: Discussionmentioning

confidence: 99%

Audio source separation into the wild

Girin

Gannot

2019

Multimodal Behavior Analysis in the Wild

View full text Add to dashboard Cite

show abstract

“…From Eq. (16) and (17), it can be seen that IRM-based approaches could deliver a less distorted enhanced speech, while it could potentially involves much interference [87]. Wang and Wang [88] first introduced DNNs to perform IBM estimation for speech separation, and reported large performance improvement over non-DNN-based methods.…”

Section: B Masking-based Deep Enhancement Methodsmentioning

confidence: 99%

“…This conclusion was further supported by the work in [90], where the obtained results suggested that IRM achieves better ASR performance than IBM. Further, motivated by the advantages and disadvantages of IBM and IRM, Grais et al [87] combined the IBM-and the IRM-based enhanced (separated) speech by another neural network, to exploit the compensation between two approaches.…”

Section: B Masking-based Deep Enhancement Methodsmentioning

confidence: 99%

Deep Learning for Environmentally Robust Speech Recognition

Zhang

Geiger

Pohjalainen

et al. 2018

ACM Trans. Intell. Syst. Technol.

285

144

View full text Add to dashboard Cite

Eliminating the negative effect of non-stationary environmental noise is a long-standing research topic for automatic speech recognition that stills remains an important challenge. Data-driven supervised approaches, including ones based on deep neural networks, have recently emerged as potential alternatives to traditional unsupervised approaches and with sufficient training, can alleviate the shortcomings of the unsupervised methods in various real-life acoustic environments. In this light, we review recently developed, representative deep learning approaches for tackling non-stationary additive and convolutional degradation of speech with the aim of providing guidelines for those involved in the development of environmentally robust speech recognition systems. We separately discuss single-and multi-channel techniques developed for the front-end and back-end of speech recognition systems, as well as joint front-end and back-end training frameworks.

show abstract

“…After obtaining the spectral weighting masks for dialogue content separation from multiple modules, these are combined with late fusion. The main task for the fusion is that it should improve the performance compared to the single best module, e.g., by locally selecting the best module based on a quality prediction [53], by using a DNN for combining the separation results [54,55], or by a weighted Table 1. Separation performance comparison of the proposed combined method (All), and the individual separation modules (Secs.…”

Section: Fusion Of Separation Modulesmentioning

confidence: 99%

Source Separation for Enabling Dialogue Enhancement in Object-based Broadcast with MPEG-H

Paulus¹,

Torcoli²,

Uhle³

et al. 2019

J. Audio Eng. Soc.

View full text Add to dashboard Cite

Dialogue Enhancement (DE) is one of the most promising applications of user interactivity enabled by object-based audio broadcasting. DE allows personalization of the relative level of dialogue for intelligibility or aesthetic reasons. This paper discusses the implementation of DE in object-based audio transport with MPEG-H, with a special focus on source separation methods enabling DE also for legacy content without original objects available. The userbenefit of DE is assessed using the Adjustment/Satisfaction Test methodology. The test results demonstrate the need for an individually adjustable dialogue level because of highly-varying personal preferences. The test also investigates the subjective quality penalty from using source separation for obtaining the objects. The results show that even an imperfect separation result can successfully enable DE leading to increased end-user satisfaction.

show abstract

Combining Mask Estimates for Single Channel Audio Source Separation Using Deep Neural Networks

Cited by 31 publications

References 29 publications

Audio source separation into the wild

Audio source separation into the wild

Deep Learning for Environmentally Robust Speech Recognition

Source Separation for Enabling Dialogue Enhancement in Object-based Broadcast with MPEG-H

Contact Info

Product

Resources

About