DescriptionRemoving silence trimming from WebRTC quality test
It turns out PESQ can deal with delays, so we don't need to
silence-trim the files to get them to line up. This solves another
problem where the silence-trimming becomes too aggressive on Windows
with the new ref file.
I also need to monitor if the new ref file, which is a lot lower
in volume than the old one, unfairly punishes the score. The
speaker volume is a bit random on the bots so the recording could
end up being quite low. Tests on my workstation suggests that PESQ
is actually pretty good about ignoring that too though. There will
be a slight effect on the score (like 0.4 MOS), which I think is
acceptable. Otherwise we need to find a way to control the speaker
volume on the bots much more stringently.
BUG=446859
Committed: https://crrev.com/36d358e0cdff201a99501e9c986a8a26f8965a40
Cr-Commit-Position: refs/heads/master@{#314797}
Patch Set 1 #Patch Set 2 : Reverting change to force_mic_volume_max; appears pesq deals with level differences #Patch Set 3 : Reverting some old changes #Patch Set 4 : #Patch Set 5 : Rebase #Messages
Total messages: 9 (3 generated)
|