To-Combo-SAD

Speech activity detection

NOTE:

Any public reports or papers derived using these scripts should cite the following two papers:

[1] S.O. Sadjadi, J.H.L. Hansen, "Unsupervised Speech Activity Detection using Voicing Measures and Perceptual Spectral Flux," IEEE Signal Processing Letters, vol. 20, no. 3, pp. 197-200, March 2013

[2] A. Ziaei, L. Kaushik, A. Sangwan, J.H.L. Hansen, "Speech Activity Detection for NASA Apollo Space Missions: Challenges and Solutions," ISCA Interspeech-2014, Paper #994, Singapore, Sept. 14-18, 2014

Running

Easier run script:

./run.sh <wavfile.wav>

Sets environment
Runs ToComboSAD
Converts to RTTM
Places output in <wavfile>.ToCombo.rttm

Example:

sox /vagrant/test2.mp3 /vagrant/test2.wav
./run.sh /vagrant/test2.wav
cat /vagrant/tocombosad_test2.rttm 
SPEAKER test2   1       0       1.98    <NA>    <NA>    speech  <NA>
SPEAKER test2   1       2.52    3.17    <NA>    <NA>    speech  <NA>
SPEAKER test2   1       6.49    3.3     <NA>    <NA>    speech  <NA>
SPEAKER test2   1       10.88   2.69    <NA>    <NA>    speech  <NA>