{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

Cauchi et al - REVERB Workshop 2014 JOINT DEREVERBERATION...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon
JOINT DEREVERBERATION AND NOISE REDUCTION USING BEAMFORMING AND A SINGLE-CHANNEL SPEECH ENHANCEMENT SCHEME Benjamin Cauchi 1 , Ina Kodrasi 2 , Robert Rehr 2 , Stephan Gerlach 1 , Ante Juki´c 2 Timo Gerkmann 2 , Simon Doclo 1,2 , Stefan Goetze 1 1 Fraunhofer IDMT, Project Group Hearing, Speech and Audio Technology, Oldenburg, Germany 2 University of Oldenburg, Department of Medical Physics and Acoustics and Cluster of Excellence Hearing4All, Oldenburg, Germany [email protected] ABSTRACT The REVERB challenge provides a common framework for the evaluation of speech enhancement algorithms in the pres- ence of both reverberation and noise. This contribution pro- poses a system consisting of a commonly used combination of a beamformer with a single-channel speech enhancement scheme aiming at joint dereverberation and noise reduction. First, a minimum variance distortionless response beam- former with an on-line estimated noise coherence matrix is used to suppress the noise and possibly some reflections. The beamformer output is then processed by a single-channel speech enhancement scheme, incorporating temporal cep- strum smoothing which suppresses both reverberation and residual noise. Experimental results show that improvements are particularly significant in conditions with high reverbera- tion times. Index Terms REVERB challenge, dereverberation, noise reduction. 1. INTRODUCTION In teleconferencing applications, voice-controlled systems and hearing aids, the recorded speech signals are often cor- rupted by both reverberation and noise, resulting in speech quality and speech intelligibility degradation, as well as de- terioration in automatic speech recognition (ASR) perfor- mance. Several algorithms have been proposed in the litera- ture to deal with these issues [1–6], but the lack of a common evaluation framework made the comparison between differ- ent approaches difficult. The REVERB challenge proposes an evaluation framework aiming to facilitate the progress of speech enhancement algorithms for noisy and reverberant environments [7]. The research leading to these results has received funding from the EU Seventh Framework Programme project DREAMS under grant agreement ITN-GA-2012-316969, from the DFG Cluster of Excellence 1077 Hear- ing4All, from a GIF grant, and from the MWK PhD Program Signals and Cognition. The system proposed in this contribution consists of a commonly used combination of a beamformer and a single- channel speech enhancement scheme. First, the multi-channel input signals are processed using a minimum variance distor- tionless response (MVDR) beamformer [8], which aims to suppress sound sources not arriving from the direction of ar- rival (DOA) of the target speaker. The noise coherence matrix in the MVDR beamformer is estimated from noise-only peri- ods, determined using a voice activity detector (VAD) [9], and the DOA of the target speaker is estimated using the multiple signal classification (MUSIC) algorithm [10,11].
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full Document Right Arrow Icon
Image of page 2
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}