| Message ID | 20180203020232.GC3063@michaelspb |
|---|---|
| State | Not Applicable |
| Headers | show |
>this breaks fate > >If the changes are intended the reference must be updated by the patch causing the changes Sorry, forgot... Now, I have some questions regarding fate tests: 1) I would like to update the fate test itself : Currently, we have : silencedetect=d=-20dB I am considering changing it to : silencedetect=n=-30dB:d=.4 The reason is that the usage would be more relevant (dB applying to noise + duration set to a consistent value for this speech sample), easier to check manually in a waveform editor, and that the coverage would be extended for the new patches (silence_start=0 + log of silence_end at end of stream). Should I first publish the patch with only the fate results changed and later on another patch to update the fate test with results changed again ? Personally, I would say a single patch with all three items (patch + fate test update + fate result update) would be clearer, but I am not familiar with ffmpeg usages, so I prefer asking... 2) I just realized today that I could fix the accuracy of silence_end too, even if it is clearly much less important compared to silence_start Do you think this should be handled by another patch, or should I better group this patch with this one as they both deal with time accuracy and affect fate results (I would rather go for the latter) ? 3) Should I prepare a new fate test to cover the new "mono" mode (patch 1/4) ? Thank you! Nicolas Gaullier
On Mon, Feb 05, 2018 at 02:20:46PM +0000, Gaullier Nicolas wrote: > >this breaks fate > > > >If the changes are intended the reference must be updated by the patch causing the changes > > Sorry, forgot... Now, I have some questions regarding fate tests: > > 1) I would like to update the fate test itself : > Currently, we have : silencedetect=d=-20dB > I am considering changing it to : silencedetect=n=-30dB:d=.4 > The reason is that the usage would be more relevant (dB applying to noise + duration set to a consistent value for this speech sample), easier to check manually in a waveform editor, and that the coverage would be extended for the new patches (silence_start=0 + log of silence_end at end of stream). > Should I first publish the patch with only the fate results changed and later on another patch to update the fate test with results changed again ? > Personally, I would say a single patch with all three items (patch + fate test update + fate result update) would be clearer, but I am not familiar with ffmpeg usages, so I prefer asking... > > 2) I just realized today that I could fix the accuracy of silence_end too, even if it is clearly much less important compared to silence_start > Do you think this should be handled by another patch, or should I better group this patch with this one as they both deal with time accuracy and affect fate results (I would rather go for the latter) ? each independant change should be in a seperate patch > > 3) Should I prepare a new fate test to cover the new "mono" mode (patch 1/4) ? if you like, yes more complete test coverage is always good thanks [...]
--- ./tests/ref/fate/filter-metadata-silencedetect 2018-02-03 00:37:40.522808340 +0100 +++ tests/data/fate/filter-metadata-silencedetect 2018-02-03 02:57:28.310985048 +0100 @@ -4,9 +4,9 @@ pkt_pts=960 pkt_pts=1280 pkt_pts=1600 -pkt_pts=1920|tag:lavfi.silence_start=0.02 +pkt_pts=1920|tag:lavfi.silence_start=0.0351875 pkt_pts=2240 -pkt_pts=2560|tag:lavfi.silence_end=0.16|tag:lavfi.silence_duration=0.14 +pkt_pts=2560|tag:lavfi.silence_end=0.16|tag:lavfi.silence_duration=0.124813 pkt_pts=2880 pkt_pts=3200 pkt_pts=3520