README FILE FOR: 2019 NIST SRE Audio-Visual Development and Test Sets
LDC Catalog-ID: LDC2023V01


1. Introduction

This package contains the SRE19 Audio-Visual Development and Test sets. 
The directory structure is organized as follows:

 docs/

  README.txt  this file
  2019_nist_multimedia_speaker_recognition_evaluation_plan_v3.pdf

  dev/        lists, keys and annotations for each partition
  eval/       (see section 2. for details)
   
 data/	      (see section 3.)

  dev/        video (mp4) files in the development set
   enrollment/   52 files
   test/        108 files

  eval/       video (mp4) files in the test set
   enrollment/  149 files
   test/        452 files


2. Overview of documentation

The "dev" and "eval" directories under "./docs/" each contain six
tab-delimited tables, as follows:

2.1  sre19_av_(dev|eval)_enrollment_boundingbox.tsv

  Col#  header_label    sample_value
  1	segmentid	engsptyj_sre19
  2	target_type	target
  3	face_frame_sec	10
  4	bounding_box	208,1,398,275
  5	face_covered	partface
  6	eyewear 	yes
  7	facial_hair	beard

2.2  sre19_av_(dev|eval)_enrollment_diarization.tsv

  Col#  header_label    sample_value
  1	segmentid	trydllly_sre19
  2	speaker_type	target
  3	start   	0.35
  4	end     	6.10

2.3  sre19_av_(dev|eval)_enrollment.tsv

  Col#  header_label    sample_value
  1	modelid 	1976_sre19
  2	segmentid	engsptyj_sre19
  3	side    	a

2.4  sre19_av_(dev|eval)_segment_key.tsv

  Col#  header_label    sample_value
  1	segmentid	engsptyj_sre19
  2	subjectid	19
  3	gender  	male
  4	partition	enrollment

2.5  sre19_av_(dev|eval)_trial_key.tsv

  Col#  header_label    sample_value
  1	modelid 	1976_sre19
  2	segmentid	afyuffzi_sre19
  3	side    	a
  4	targettype	nontarget

2.6  sre19_av_(dev|eval)_trials.tsv

  Col#  header_label    sample_value
  1	modelid 	1976_sre19
  2	segmentid	afyuffzi_sre19
  3	side    	a


3. Overview of video data

All videos are encoded/compressed as "mp4" files.  Audio tracks all have a
sample rate of 44.1 KHz with AAC encoding; most are stereo, but some are mono.  File durations
range from 17.5 seconds to nearly 13 minutes.

The audio can be extracted for processing using ffmpeg as follows:

  ffmpeg -v 8 -i <video_in> -vn -ar 16000 -ac 1 -f wav <wav_out>

Here, the audio is extracted at 16 kHz, but the videos do support higher/lower 
rates for processing if desired.

Specific frames (in seconds) can be extracted from videos using ffmpeg as follows:

  ffmpeg -v 8 -ss <face_frame_sec> -i <video_in> -vframes 1 <image_out>

The following ffmpeg command can be used to extract a frame every second:

  ffmpeg -v 8 -i <video_in> -vf fps=1 <image_out>_%04d.png


-------------
README originally created by Omid Sadjadi, July 19, 2019
 adapted for general release by David Graff, July 1, 2021