Community Research and Development Information Service - CORDIS

Final Report Summary - DREAMS (Dereverberation and Reverberation of Audio, Music, and Speech)

The DREAMS Initial Training Network (ITN) has investigated the problem of modeling, controlling, removing, and synthesizing acoustic reverberation with the aim of enhancing the quality and intelligibility of audio, music, and speech signals.

The core activities of the DREAMS ITN are research and training, and have been carried out by a consortium of 4 academic and 8 industrial partners located in Belgium, Denmark, Germany, and the UK. The ITN has been hosting 12 early stage researchers and 4 experienced researchers, each performing an individual research project around one of four themes that reflect the most challenging open problems in the area of (de-)reverberation. The ITN has been implemented such as to maximize the international and intersectoral experience of the research fellows, by defining relevant secondments in academia and industry, both in the host country and abroad.

The research training program of the DREAMS ITN revolves around four scientific disciplines that are equally important in understanding and tackling the (de-)reverberation problem: room acoustics, psychoacoustics, speech and audio processing, and signal processing. The DREAMS ITN has taken a significant step forward and made fundamental scientific contributions in each of these four disciplines:

(1) Our objective regarding room acoustics consisted in the development of new room acoustics models and the establishment of interconnections between existing models. The most important results obtained are the identification of perceptually relevant attributes for small room and car cabin acoustics, the development of perceptually meaningful approximations to physical room acoustics models, and the design of accurate, scalable, and blind estimation algorithms for microscopic as well as macroscopic room models.

(2) In the area of psychoacoustics, we aimed to quantify the perceptual impact of reverberation and dereverberation on speech and audio signals, and to steer the design of dereverberation algorithms accordingly. Key results obtained are an extensive evaluation and rating of objective measures for the intelligibility and quality of dereverberated speech, new experimental paradigms and metrics to assess the perceptual impact of reverberation, and novel design methodologies for single- and multi-channel speech enhancement.

(3) In terms of speech and audio processing, our objective has been to design algorithms that are capable of compensating reverberation combined other acoustic artefacts such as noise, echo, coding artefacts, and nonlinear distortion. Our research progress in this area includes the development of sparsity-promoting multi-channel algorithms for joint dereverberation and echo cancellation and joint dereverberation and noise reduction, as well as information theoretic results and coding schemes for data compression and enhancement of reverberant audio signals.

(4) In the area of signal processing, our objective has been to develop numerically robust and computationally efficient algorithms for dereverberation and related inverse acoustic problems. Key results include novel algorithms for sound source localization, sound field control, dereverberation, and room impulse response interpolation based on numerical room acoustics models, multi-channel acoustic system inversion robust to noise and under-modelling, and an adaptive beamforming framework for dereverberation.

Next to these fundamental scientific challenges, the DREAMS ITN has defined a number of technological objectives which are strongly application-driven:

(1) Personal audio devices: with the aim of obtaining maximally transparent audio quality at a reasonable computational cost, we have been developing efficient speech and audio enhancement algorithms based on scalable room acoustics models .

(2) Hearing aids: to obtain significant improvements in speech intelligibility for hearing aid users in adverse environments, novel multi-channel dereverberation and joint dereverberation and noise suppression algorithms have been designed and evaluated.

(3) Public address systems: reverberation control for improving the sound quality of speech and audio reproduction in public address systems has been targeted by developing efficient and perceptually relevant multi-channel room acoustics models and equalization algorithms.

(4) Automatic speech recognition (ASR): advances in reverberation-robust ASR have been achieved by combined front-end and back-end processing based on blind estimation of the speech clarity index (C50), which has been shown to be highly correlated to ASR accuracy.

(5) Acoustic monitoring and surveillance: to quantify and improve the speech intelligibility of low-quality recordings obtained with distributed microphones, we have developed an improved psychometric function estimation procedure, as well as coding schemes for distributed reverberant audio compression.

With the aim of equipping the DREAMS ITN research fellows with the best possible skills to carry out their research projects and develop their careers, a high-quality training program has been implemented, with local as well as network-wide training modules. The network-wide training events include four DREAMS Seasonal Schools, during which relevant technical topics have been addressed in guest lectures by internationally leading experts, e.g. Reverberation Perception (by Barry Blesser), Electroacoustics (by Peter Swarte), Room Acoustics (by Diemer de Vries), Spatiotemporal Acoustic Signal Processing (by Walter Kellermann), Adaptive Filters, Semi-Supervised Learning and Cognitive Systems (by Simon Haykin), Binaural Hearing and Communication Acoustics (by Jens Blauert), Machine Learning (by Sergios Theodoridis), Compressive Sensing (by Wei Dai), and Deep Neural Networks (by Kate Knill). In addition, these Seasonal School featured hands-on application sessions as well as non-technical courses on Team Building, Marketing & Entrepreneurship, Intellectual Property Rights (IPR) Management, Reproducible Research, Project Proposal Writing, Negotiation Skills, Career Development, Productivity Enhancement, and Leadership Skills. Apart from the DREAMS Seasonal Schools, a network-wide Workshop on The Impact of Reverberation and Noise on Speech Intelligibility and Speech Recognition has been held (with keynote lectures by Gaston Hilkhuysen, Stefan Goetze, and Jan Rennies) and a PhD School on Advanced Psychoacoustics: Theory, Practice, and Modeling (with guest lectures by Dorte Hammershøi, Steven van De Par, Jasper van Dorp Schuitman, Jan Larsen, Wolfgang Ellermeier, Nick Zacharov, Søren Bech, Alexander Raake, Alexander Lindau, and Tapio Lokki) has been organized jointly with the European Association for Signal Processing (EURASIP).

The DREAMS ITN has organized the Audio Engineering Society (AES) 60th International Conference on Dereverberation and Reverberation of Audio, Music, and Speech, in Leuven, Belgium, Feb. 3-5, 2016 (http://www.aes.org/conferences/60/). The aim of this international conference was to bring together and provide a forum for researchers working on different aspects and applications of reverberation and dereverberation. In addition, the DREAMS ITN has organized special sessions at several international conferences, and a recently published special issue of the Journal of the Audio Engineering Society (http://www.aes.org/journal/online/JAES_V65/1_2/).

Finally, the DREAMS ITN has been committed to communicate its research progress not only to the scientific community but also to the public-at-large. To this end, a broad range of outreach activities has been organized, such as presenting hands-on demonstrations and fellows’ testimonies in local science communication events, the promotion of the DREAMS ITN research topics on social media, and the organization of live events related to reverberation and music. The largest outreach activity took place in July 2015, with the participation of the DREAMS ITN in the Summer Science Exhibition in London, a prestigious science communication event organized by the Royal Society that attracted over 13000 visitors (http://sse.royalsociety.org/2015/sound-interactions/).

Project website: http://www.dreams-itn.eu/

Related information

Reported by

KATHOLIEKE UNIVERSITEIT LEUVEN
Belgium

Subjects

Life Sciences
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top