Beginner problem with Silence Recognizer
Up one level
Beginner problem with Silence Recognizer
I just started using ELAN a couple days ago. I have a video and sound file that were recorded in a classroom setting (with associated ambient noise of backpack zippers, chalk hitting blackboard, paper shuffling) and I was trying to use the silence recognizer to get a segmented tier. <br><br>I gave about 10 examples of silence to the application and hit Start. It put one S tag at the beginning of the waveform. And that's it. Here is the Report:<br><br>Noise level 1: 16860.68<br>Minimal silence duration : 202 ms<br>Minimal non-silence duration : 199 ms<br>Number of segments channel 1: 1<br><br>Processing took: 15884 ms<br><br><br><br>Could someone tell me what the problem might be? Even if the ambient noise sometimes makes a higher amplitude waveform than the speech itself, the software should find at least a few of *obvious* silences. <br><br>Thanks!<br>Neiloufar<br>
Re: Beginner problem with Silence Recognizer
Hi, the classic silence recognizer is only for PERFECT silence as far as I remember. Such silence does not even contain microphone noise. Sometimes it is still useful to find timespans where e.g. the microphone was switched off or where sound was removed during editing. Avatech (www.mpi.nl/avatech) also has recognizers for relative silence and other things. It should not be necessary to give any examples for silence. Due to potential licensing issues, you might want to ask Han to add a threshold (e.g. in decibels) to let the classic recognizer also find relative silence, in case you cannot use the newer Avatech recognizers. For details about Avatech, you can email: eric.auer X mpi.nl (X = @-sign).
Re: Beginner problem with Silence Recognizer
The situation with the Silence Recognizer is not quite like that; it does take the average energy of the provided examples into account to detect silence versus non-silence segments. It does not, however, distinguish between speech and non-speech noise.
So, could it be you selected some examples with a lot of (high amplitude) noise instead of segments containing only the background noise (which is something different than absolute silence)? The noise level in your report ("Noise level 1: 16860.68") seem awfully high, suggesting you provided some high volume noise examples. It is probably best to select just a few segments with relatively low noise.
-Han
Release of ELAN 4.3.0
