Multispeaker detection-localization: an implementation

The corresponding research report is IDIAP RR 06-26.

Unless you are on an Intel Pentium (4) platform, you may need to recompile the MEX files. Do not worry, this is quite easy. See the instructions at the top of FAST_detect_locate.m. If you still cannot run FAST_detect_locate with the MEX files, you can try it with the original MATLAB implementation activated (look for if 1 in FAST_detect_locate.m). Of course this would be much slower.

Look at README.m if you want to understand each step, and/or you are interested in tests with multiple moving/seated speakers.
Look at MEETING_SEG.m if you are only interested in meeting segmentation (subset of README.m).
Both rely on the same implementation.

Compatibility issues: there are 3 binary data files in the "DATA" subdirectory, that were created with MATLAB 6.5.1. If your MATLAB version cannot read those, then use the 3 companion MATLAB functions to recreate them ("DATA/*_mat.m" ASCII files).

Questions? ->

Back to the audio resources page

/com/mmm/shared/lathoud/2006-multidetloc/

2006-multidetloc-code.tar.gz [archive of this directory (1.9 M)]

2006-multidetloc-data.tar.gz [archive of the DATA subdirectory (300 M)]

DATA/ [all eight recordings, their ground-truth, and the calibration file]

FASTTDE_detect_locate.m [main MATLAB function, "FASTTDE" implementation (=faster but imprecise).]

FASTTDE_detect_locate_wrapper.m

FAST_detect_locate.m [main MATLAB function, "FAST" implementation]

FAST_detect_locate_wrapper.m

FULL_detect_locate.m [main MATLAB function, "FULL" implementation (=slow).]

FULL_detect_locate_wrapper.m

MEETING_SEG.m [a stripped down version of README.m, if you only need to segment meetings]

README.m [where you should start]

angle_within.m

cart2sph_mat.m

check_2d2g1se_model.m

check_param.m

chrono_check.m

chrono_start.m

chrono_stop.m

classify_cluster_sns.m

classify_frame_sns.m

close_seg.m

compute_gccphat_c.c

compute_gccphat_c.mexglx

compute_post_2d2g1se.m

compute_post_gmm.m

compute_post_raylsherl.m

compute_smooth_loctimeseg.m

construct_raw_loctimeseg.m

define_silence_in_seg.m

determinit_gmm.m

dE_E_fun.m [part 1/2 of the MATLAB implementation equivalent to the MEX C implementation onearray_run_scg_fast_logsph_c.c]

dilate_seg.m

dilation.m

dilation_rows.m

display_cost.m

display_raw.m

display_raw_loop.m

display_stats.m

display_stats_sbmfcc.m

do_em_gmm.m

erode_seg.m

estimate_K_gmm.m

eval_detloc.m

example_cluster_level_sns.m

example_frame_level_sns.m

far_frr.m

fill_default.m

fit_raylsherl.m

get_ssm_activeness_c.c

get_ssm_activeness_c.mexglx

index.html.save

index.html~

location_to_timedelay.m

log_mvnpdf.m

log_normpdf.m

main_loop.m

main_loop_gmm.m

my_logsum_fast.m

my_logsum_fast.pdf

one_array_fasttde_dirloc_c.c

one_array_fasttde_dirloc_c.mexglx

onearray_run_scg_fast_logsph_c.c

onearray_run_scg_fast_logsph_c.mexglx

open_seg.m

pair_distance.m

peigne_vec.m

plot_loctimeseg.m

plot_multi_cluster.m

run_scg_fast.m [part 2/2 of the MATLAB implementation equivalent to the MEX C implementation onearray_run_scg_fast_logsph_c.c]

sbmfcc_sns_model.m

seq11-1p-0100-FAST-before-SNS.eps

seq11-1p-0100-FAST.eps

seq18-2p-0101-FAST.eps

seq18-cam2.jpg

seq18.jpg

seq24-2p-0111-FAST.eps

seq40-3p-0111.eps

seq45-3p-1111-FAST.eps

seq45-cam2.jpg

seq45.jpg

show_static_seg.m

simplify_seg.m

sph2cart_mat.m

static_bic.m

static_sigma.m

train_gmm_bic.m

wa_sns_model.m