AlexandrosTsilfidis

[D1] A. Tsilfidis, "Signal Processing Methods for Enhancing Speech and Music Signals in Reverberant Environments [pdf]" (under completion)

2. Journal Papers

[J2] A. Tsilfidis and J. Mourjopoulos, "Blind single-channel suppression of late reverberation based on perceptual reverberation modeling" [pdf] , Journal of the Acoustical Society of America, 129(3), p. 1439-1451, 2011. (Copyright 2011 Acoustical Society of America. This article may be downloaded for personal use only. Any other use requires prior permission of the author and the Acoustical Society of America. JASA link: DOI: 10.1121/1.3533690)

[J1] A. Tsilfidis and J. Mourjopoulos, “Signal-dependent constraints for perceptually motivated suppression of late reverberation” [pdf] , Signal Processing, Volume 90, pp. 959-965, 2010. (Elsevier Link: http://dx.doi.org/10.1016/j.sigpro.2009.09.020)

3. International Conference Proceedings

[C11] A. Tsilfidis , T. Vovolis , E. Georganti, P. Teubner, J. Mourjopoulos, "Acoustic radiation properties of ancient greek theatre masks"[pdf], The Acoustics of Ancient Theatres Conference, Patras, Greece.

[C10] A. Tsilfidis, E. Georganti, E. K. Kokkinis and J Mourjopoulos, "Speech dereverberation based on a recorded handclap",Digital Signal Processing Conference (DSP), Corfu, Greece, 2011.

[C9] A. Tsilfidis, E. Georganti and J. Mourjopoulos, "A binaural framework for spectral subtraction dereverberation", Forum Acusticum (invited paper), Aalborg, Denmark, 2011.

[C8] A. Tsilfidis, E. K. Kokkinis and J. Mourjopoulos, "Suppression of late reverberation at multiple speaker positions utilizing a single impulse response measurement" Forum Acusticum, Aalborg, Denmark, 2011.

[C7] A. Tsilfidis, E. Georganti and J. Mourjopoulos, "Binaural extension and performance of single-channel spectral subtraction dereverberation algorithms" [pdf] , IEEE ICASSP, Prague, Czech Republic, 2011.

[C6] A. Tsilfidis and J. Mourjopoulos, "Blind single-channel dereverberation for music post-processing" [pdf], 130th Convention of the Audio Engineering Society, London, UK, 2011

[C5] E. K. Kokkinis, A. Tsilfidis, E. Georganti and J. Mourjopoulos, "Joint noise and reverberation suppression for speech applications" [pdf], 130th Convention of the Audio Engineering Society, London, UK, 2011.

[C4] E. Georganti, A. Tsilfidis and J. Mourjopoulos, "Statistical Analysis of Binaural Room Impulse Responses" [pdf], 130th Convention of the Audio Engineering Society, London, UK, 2011.

[C3] A. Tsilfidis and J. Mourjopoulos, "Perceptually-motivated selective suppression of late reverberation", 6th International Conference on Digital Signal Processing, Santorini, Greece, July 2009.

[C2] A. Tsilfidis, C. Papadakos and J. Mourjopoulos, "Hierarchical Perceptual Mixing" [pdf], 126th Convetions of the Audio Engineering Society, Munich, Germany, May 2009.

[C1] A. Tsilfidis, J. Mourjopoulos, D. Tsoukalas, “Blind Estimation and Suppresion of late reverberation utilising auditory masking” [pdf] presented at “Hands-Free Speech Communication and Microphone Arrays 2008”, Trento Italy, May 2008.

4. Greek Conference Proceedings

[3] A.Tsilfidis "Auditory thresholds of Gaussian Shaped Sinusoidal Tones", Hellenic Institute of Acoustics 2008 conference (in Greek), Xanthi, September 2008.

[2] A. Tsilfidis, J. Mourjopoulos, D. Tsoukalas, "Method for estimation and suppression of reverberation using psychoacoustic criteria", Hellenic Institute of Acoustics 2008 conference (in Greek), Xanthi, September 2008.

[1] A. Tsilfidis and J. Mourjopoulos, "Blind Dereverberation for Speech and Music Signals", Hellenic Institute of Acoustics 2010 conference (in Greek), 2010.

1. Blind Single Channel Dereverberation

This is a blind dereverberation method based on perceptual reverberation modeling. This technique employs a computational auditory masking model and locates the signal regions where late reverberation is audible, i.e. where it is unmasked from the clean signal components. Following a selective signal processing approach, only such signal regions are further processed through sub-band gain filtering and therefor annoying processing artifacts are minimized. The technique has been evaluated for both speech and music signals and for a wide range of reverberation conditions. In all cases it was found to produce perceptually superior clean signal estimations than any other tested technique. Moreover, extensive Automatic Speec Recognition tests have shown that it significantly improves the recognition performance, especially in highly reverberant environments. The the related publications are: [J2] and [C6]

These are audio examples of speech and music in a measured Concert Hall (RT60=1.47 sec). The source-microphone distance is 1.5m.

Reverberant Speech signal        Dereverberated Speech signal   

Reverberant Cello signal        Dereverberated Cello signal   

Reverberant Orchestra signal        Dereverberated Orchestra signal   

And these are examples of an acoustic guitar in a measured Lecture Hall (RT60=1 sec). The source-microphone distance is again set at 1.5m.

Reverberant Guitar signal        Dereverberated Guitar signal   

2. Blind Binaural Dereverberation

Apart from the challenging task of reducing reverberation without introducing audible artifacts, binaural dereverberation methods should also at least preserve the Interaural Time Difference (ITD) and Interaural Level Difference (ILD) cues as it has been shown that bilateral signal processing affects the source localization. These sound samples demonstrate a generalized framework for binaural spectral subtraction dereverberation. The proposed approach is based on the binaural extension of state-of-the-art single-channel late reverberation suppression techniques and relies on bilateral gain adaptation, a technique which efficiently reduces reverberation and also preserves the binaural localization cues. Moreover, a Gain Magnitude Regularization (GMR) step is implemented in order to reduce overestimation artifacts. The related publications are: [C7] and [C9]

Please listen to these demos with headphones!

The reverberant signal has been generated through convolution with a measured Binaural Impulse Response. The response was measured in a cafeteria (RT=1.25 sec, Dist.=1.18 m, azim = -30) taken from the Oldenburg database .

Reverberant       

Listen now to some dereverberation approaches based on different gain adaptation techniques:

Adaptation	LB	WW	FK
mingain	Dereverberated	Dereverberated	Dereverberated
avggain	Dereverberated	Dereverberated	Dereverberated
maxgain	Dereverberated	Dereverberated	Dereverberated

3. Semi-blind Dereverberation for Speech

We have also proposed an efficient spectral subtraction dereverberation method that estimates the late reverberation’s spectral magnitude utilizing the late part of a measured impulse response and the excitation signal derived from the Linear Prediction (LP) analysis of the reverberant speech signal. Moreover, we have extended the above technique in a semi-blind framework utilizing a single recorded handclap. This is a flexible option when a RIR measurement is not feasible since a handclap recording can provide a reasonable RIR approximation. The related publications are: [C8], [C10] and [C5]

Reverberant speech        Dereverberated speech   

UPDATE: An improved version of the algorithm has been submitted for publication. Check the videos below for dereverberation demos in real moving speaker scenarios!
The first recording was made in a classroom with a reverberation time of approximately 1 second. The room's air conditioning unit was operating during the recording.



The second recording was made in a classroom with a reverberation time of approximately 2 seconds. The room's air conditioning unit was operating during the recording. This video demonstrates the method's performance in adverse acoustical conditions.



The dataset (real measurements) for our moving speaker experiments is available under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.

4. Hierarchical Perceptual Mixing

This is a method for perceptually-motivated signal dependent audio mixing. The proposed Hierarchical Perceptual Mixing (HPM) method is implemented in the spectro-temporal domain; its principle is to combine only the perceptually relevant components of the audio signals, derived after the calculation of the minimum masking threshold which is introduced in the mixing stage. The resulted signals have enhanced dynamic range and lower crest factor with no unwanted artifacts, compared to the traditionally mixed signals. Moreover, the overall headroom is improved, while clarity and tonal balance are preserved. The related publications is: [C4] .

This demo has been made by mixing of 3 input sounds: 2 channels of electric guitar and 1 channel vibraphone. In the first example we are using the guitar as a dominant sound:

Conventional Mix        HPM using a guitar as dominant sound    Difference: HPM - Conventional Mix (6 dB boost)   

while in the second example the vibraphone has been used as a dominant sound:

Conventional Mix        HPM using a vibraphone as dominant sound    Difference: HPM - Conventional Mix (6 dB boost)   

The proposed technique can be also used to create some creative audio effects. Listen an example of how to use HPM for acoustic guitar dubbing

L-R dubbing of 2 acoustic guitars        Mono mix of 2 acoustic guitars    Effect produced when panning the HP mixes of the above guitars