Online resources for audio processing
by Guillaume Lathoud
These are mostly data (video and audio), and MATLAB/C code for microphone array speaker localization,
as well as single microphone noise/channel removal.
Feel free to use any of these resources, as long as you properly cite the corresponding papers
in your own publications and implementations.
To my home page.
To the AV16.3 Corpus home page.
[ DATA ]
[ MATLAB & C code ]
The AV16.3 Corpus for Audio-Visual Speaker Localization and Tracking.
(two microphone arrays, three cameras, 3-D mouth location known within 1.2 cm).
Includes MATLAB code, annotation GUIs and examples (2004 MLMI paper).
2005-SAM-SPARSE-MEAN.tar.gz (whole directory: 141 M)
MATLAB and MATLAB/C implementations of the sector-based detection-localization with microphone array
(ICASSP 2005, EURASIP 2006 and IDIAP RR-04-67, by G. Lathoud and others).
2006-CHN-USS.tar.gz (whole directory: 16 M)
MATLAB implementation of single microphone noise & channel removal (IDIAP RR-06-09).
2006-CHN-USS-TIME-DOMAIN.tar.gz (whole directory: 269 k)
MATLAB implementation of single microphone noise & channel removal (IDIAP RR-06-09)
+ reconstruction of the cleaned waveform through overlap-add.
2006-short-term-clustering.tar.gz (whole directory: 2.8 M)
MATLAB implementation of short-term clustering (ML & confident),
along with simple examples on synthetic data,
including threshold-free detection of trajectory crossings.
2006-multidetloc-code.tar.gz (whole directory, first part: 1.9 M)
2006-multidetloc-data.tar.gz (whole directory, second part: 308 M)
MATLAB/C implementation of multispeaker detection-localization
applied to real microphone array recordings of multiple humans (AV16.3 Corpus)
(this demonstrates short-term clustering on real data)
+ evaluation of the result against the 3-D location ground-truth
2006-distant-speaker-code.tar.gz (whole directory, first part: 183 k)
2006-distant-speaker-data.tar.gz (whole directory, second part: 194 M)
MATLAB cross-correlation study on speech signals received by distant microphones.
Signals are emitted by real human speakers (seq01 & seq03 in the AV16.3 Corpus)
(corresponding paper: IDIAP RR-06-74).
2006-various-MATLAB-tools.tar.gz (whole directory: 84 k)
That's what it says!
MULTISEG.tar.gz (whole directory: 28 M)
Multispeaker speech/silence segmentation using multiple lapel microphones.
This is a MATLAB implementation of the lapel segmentation baseline in the 2004 NIST Workshop paper.
USS-EXAMPLE.zip (whole directory: 11 M)
MATLAB implementation of the single channel noise removal procedure
called "Unsupervised Spectral Subtraction" (ASRU 2005, IDIAP RR-05-42).
Wiener_standalone.tar.gz (whole directory: 790 k)
MATLAB implementation of Wiener filtering in the MEL domain (one or multiple iterations).
A Python wrapper is also given.
Last updated on 2008-09-19 by Guillaume Lathoud - glathoud at yahoo dot fr