You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
ffmpeg does not support multiple filter arguments, so the highpass filter is not applied. To be effective, need to be chained eg highpass=frequency=1000,silencedetect=noise=0.15:duration=1
more importantly, these settings are very different from what the old silan-based detection used.
Silan used a threshhold of 0.001 (equivalent to -60dB), 0.15 as used now is equivalent to -16dB. This is very aggressive, which is especially noticeable with spoken-word programming as final syllables or words can get detected as silence.
1000Hz for a cutoff seems incredibly high (although this has no impact as it's not currently being applied) I'm not certain how silan's high pass coefficient maps to a frequency, but the 0.99 value Libretime used seems to be equivalent to about 80Hz which is a reasonable value.
To be consistent with the silence detection before the change to ffmpeg was implented, the filter should be something like: highpass=frequency=80,silencedetect=noise=0.001:duration=1
To reproduce
Upload a file, and review cue points
In audio editing software, confirm that cue points are at logical place
Expected behavior
As silence detection, cue in and out points should not trim off any of the content of the track.
For example, here is the results for the cue-out point identified for a track in our music library:
The yellow marker is the current behaviour of libretime_analyzer, which trims off a bit of the end of the track.
The red marker marker is the cue point ffmpeg reports if the highpass filter at 1000Hz is applied, trimming things even more aggressively.
The green marker is the cue point ffmpeg reports with the values suggested to be consistent with the old silan behaviour, which does not trim off any actual audio in the track.
Hey @rjhelms,
Haha, indeed, I might have been a bit aggressive on the settings here.
And I didn't know we couldn't chain -filter arguments, thanks for brigging this up.
It was my intention to make this configurable from the UI, but the analyzer doesn't support passing configuration from the UI/legacy app yet, we might be able to use the configuration file, but I think this would be a bad idea.
I have some branches working on this locally, never had the chance to push it. I'll see if I can push some change to make this happen.
This issue has been automatically marked as stale because it has not had activity in the last 5 months. It will be closed if no activity occurs in the next month.
Please chat to us on the forum or ask for help on #libretime:matrix.org if you have any questions or need further support with getting this issue resolved.
You may also label an issue as pinned if you would like to make sure that it does not get closed by this bot.
Describe the bug
The call to run ffmpeg for detect silences is as follows:
libretime/analyzer/libretime_analyzer/pipeline/_ffmpeg.py
Lines 75 to 80 in bd6822f
There are two issues with this call:
highpass=frequency=1000,silencedetect=noise=0.15:duration=1
silan
-based detection used.To be consistent with the silence detection before the change to ffmpeg was implented, the filter should be something like:
highpass=frequency=80,silencedetect=noise=0.001:duration=1
To reproduce
Expected behavior
As silence detection, cue in and out points should not trim off any of the content of the track.
For example, here is the results for the cue-out point identified for a track in our music library:
Relevant log output or error messages
No response
LibreTime version
3.0.1
Installation method and OS / Environment
Operating system: Ubuntu 20.04.5 LTS
Method: install script
Installation details
No response
Client Environment
No response
Screenshots
No response
The text was updated successfully, but these errors were encountered: