michaelni changed the topic of #ffmpeg-devel to: Welcome to the FFmpeg development channel | Questions about using FFmpeg or developing with libav* libs should be asked in #ffmpeg | This channel is publicly logged | FFmpeg 7.1 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
bencoh has quit [Ping timeout: 260 seconds]
bencoh has joined #ffmpeg-devel
bencoh has quit [Ping timeout: 244 seconds]
bencoh has joined #ffmpeg-devel
Marth64 has quit [Quit: Leaving]
bencoh has quit [Ping timeout: 260 seconds]
bencoh has joined #ffmpeg-devel
HarshK23 has quit [Quit: Connection closed for inactivity]
iive has quit [Quit: They came for me...]
thilo has quit [Ping timeout: 252 seconds]
thilo has joined #ffmpeg-devel
cone-753 has quit [Quit: transmission timeout]
averne has quit [Ping timeout: 252 seconds]
averne_ has joined #ffmpeg-devel
averne_ is now known as averne
<mkver>
Clang-ASAN reports a stack buffer overread in checkasm-sw_rgb here (in check_yuv2packed1, yuv2rgb_1_c_template).
<fflogger>
[newticket] steven1: Ticket #11407 ([undetermined] heap-buffer-overflow vulnerability find in in function mov_read_trun at /ffmpeg/libavformat/mov.c:5944) created https://trac.ffmpeg.org/ticket/11407
<mkver>
Seems like I sent a patch twice; sorry for that.
mkver has quit [Ping timeout: 252 seconds]
^Neo has quit [Ping timeout: 276 seconds]
jamrial has quit []
System_Error has quit [Remote host closed the connection]
System_Error has joined #ffmpeg-devel
Martchus has joined #ffmpeg-devel
Martchus_ has quit [Ping timeout: 248 seconds]
* Compn
facepalms
HarshK23 has joined #ffmpeg-devel
ngaullier has joined #ffmpeg-devel
Thulinma has quit [Read error: Connection reset by peer]
<mkver>
If one uses (int16_t *)src_u + i*INPUT_SIZE, the stack-buffer overflow is fixed; but I don't get why you use int32_t for data that is only treated as int16_t later (this is a effective-type violation and therefore UB).
<mkver>
Is it because it simplified initialization?
ngaullie has quit [Remote host closed the connection]
ngaullier has joined #ffmpeg-devel
ngaullier has quit [Remote host closed the connection]
Everythi1g has joined #ffmpeg-devel
Everything has quit [Quit: leaving]
Everythi1g has quit [Client Quit]
Everything has joined #ffmpeg-devel
meego has joined #ffmpeg-devel
<meego>
Quick call for freelancer: a project at my company could really use a faster loudnorm filter (see https://ffmpeg.org/ffmpeg-filters.html#loudnorm). Current implementation is capped at x≈20 faster than realtime, but some other implementations go as high as x230. We pay well and fast, and will offer bonuses if x100 and x200 milestones are achieved.
<Compn>
ooo
<Compn>
meego, wait around here for a devel or drop your contact info if you like
<Compn>
(though this channel is publicly logged)
<Compn>
we also have trac.ffmpeg.org if you'd prefer putting your bounty/job in a more solid place than irc. also can post to ffmpeg-devel mailing list if you wish
<Compn>
the ffmpeg-devel mailing list is the preferred and most read place for developers in this project.
<Compn>
send a mail to ffmpeg-devel@ffmpeg.org and i can just approve it for you. i think it should work
<Compn>
(we're not all on irc)
<meego>
Compn: thanks for the advice. I'm reachable at tech@spectre-music.com but i'll stick around here. I wasn't sure if the ML was the right place for this.
<Compn>
no problem :)
<meego>
I'll fill a bug on trac and an email on ffmpeg-devel
<Compn>
its all just options, whichever you like :)
ngaullier has quit [Remote host closed the connection]
System_Error has quit [Remote host closed the connection]
System_Error has joined #ffmpeg-devel
ccawley2011__ has joined #ffmpeg-devel
ccawley2011_ has quit [Ping timeout: 245 seconds]
NanananaLeader has joined #ffmpeg-devel
ccawley2011__ has quit [Ping timeout: 248 seconds]
<another|>
NanananaLeader: seen !11408 ?
<NanananaLeader>
loudnorm implementation is more than 100x realtime in librempeg, the dynamic mode
<another|>
there was a bug bounty offered
<NanananaLeader>
scanning + volume pass is also > 300x faster (true peak) or > 690x (sample peak) (>1000x none peak), so for one pass just use ebur128 filter + volume filter, but this needs manual intervention
<NanananaLeader>
or scripting
<NanananaLeader>
i doubt the bounty is about dynamic loudnorm mode
<NanananaLeader>
so user is probably unaware of ebur128 + volume way
<another|>
<meego> Quick call for freelancer: a project at my company could really use a faster loudnorm filter (see https://ffmpeg.org/ffmpeg-filters.html#loudnorm). Current implementation is capped at x≈20 faster than realtime, but some other implementations go as high as x230. We pay well and fast, and will offer bonuses if x100 and x200 milestones are achieved.
<Marth64>
i always found the manual intervention for second pass cumbersome
<NanananaLeader>
the biggest slowdown in loudnorm is truepeak (aka resampling to 192000 sample rate and that is done two times, for both input and output samples)
mkver has quit [Ping timeout: 252 seconds]
<NanananaLeader>
Marth64: its not about manual, one could easily code something/script to extract values from ebur128 scan and apply it in 2nd pass with volume filter
ccawley2011__ has joined #ffmpeg-devel
<NanananaLeader>
because it need to scan all audio from start to the end there is not single line way to accomplish the task
<NanananaLeader>
but maybe bounty is really about dynamic mode in loudnorm, need to ask user...
<Compn>
the real issue is that film production companies are mixing these films with mumbling whispering dialog and 120db explosions. its insanity
<Compn>
and also tv speakers have gotten worse
mkver has joined #ffmpeg-devel
<NanananaLeader>
arent dialogs mixed into C channel?
<NanananaLeader>
aka front center
<nevcairiel>
usually yes
<Compn>
sometimes
<Compn>
but also things are mixed into that area so you cant just blanket increase gain there
<Compn>
and better speakers wont help because you're still stuck with 120db noises. but now you're throwing 120db through some powerful speakers and its shaking the windows.
<NanananaLeader>
toggle voice activity detector
<Compn>
going to a movie theater is like an assault on your ears :(
<Marth64>
NanananaLeader: I have such python script but my meaning is that its a redundant task, which it can be handled within filter itself (for example output to file then read from file as options)
<NanananaLeader>
Marth64: yea, even scripting might not be necessary, just pass some options into another pass but within ffmpeg - could be coded like this filtering the encoded streams...
<NanananaLeader>
but as ffmpeg works with streams and not only files its problematic
ccawley2011_ has joined #ffmpeg-devel
ccawley2011__ has quit [Ping timeout: 248 seconds]
<Compn>
NanananaLeader, next gen ai word detection filter, only adds gain per word
<NanananaLeader>
its more than than, arround 600x for float input
<NanananaLeader>
dunno why you force double floating processing in your usage
<Compn>
lots of floats
<jamrial>
i paid for 64 bit registers, and i'm going to use all 64 bits
<Compn>
send patch for filter in asm ?
<NanananaLeader>
why 96? bit registers are not supported ?
<jamrial>
x87 has eighty something bits intermediates
<NanananaLeader>
Compn: that filter uses fft/rdft and that already have asm SIMD for power 2 size in ffmpeg
<NanananaLeader>
long double sinl()
<NanananaLeader>
sinl() are part of posix
<NanananaLeader>
that means we must add long double sample format to libavutil/avfilter
<NanananaLeader>
and add support for such sample formats in *all* audio filters
<NanananaLeader>
patch welcome
<NanananaLeader>
jamrial: its 88 bits iirc
<NanananaLeader>
Compn: why you are unilateral supporter of leader?
<Compn>
dont support anyone with unilateral
<Compn>
thanks for asking instead of assuming
<Marth64>
lets leave people alone and stick to chat about ff tech, its more intellectual and interesting
<NanananaLeader>
lets talk about SDR then?
<NanananaLeader>
or why not moved to github/gitlab?
<Marth64>
yes i think many people want to move
<NanananaLeader>
or about other problems, like subtitles support/API etc...
<Marth64>
subtitle problems are real, I am working on a powerpoint to suggest improvements
<Marth64>
will share with you for opinions
<Marth64>
current subtitle support sucks
<NanananaLeader>
just make subtitles stream-lined aware
<NanananaLeader>
currently all subtitles are read in one go
<Marth64>
yes and I want to add filtering finally
<Marth64>
also closed captions need work. i am almost done with my improvement proposal. i am curious to your input
<Marth64>
for CC
<Marth64>
(eia608)
<NanananaLeader>
yes, readeia608 should be ultimately V->S filter not what it is now
<Marth64>
i will have something material to share by sunday for closed captions (powerpoint 1)
<Marth64>
will definitely loop you
<NanananaLeader>
what you (gonna)change with CC ?
<NanananaLeader>
if there is something very complex or new media type or what not its less likely to over be done/merged
<Marth64>
I want to propose we have a CBS at the heart of parsing the byte stream and abstract bespoke logic from various areas
<Marth64>
I will send you a draft now actually
<Marth64>
for the deck
<NanananaLeader>
also if you gonna invent new bitstream inplace of ASS that will be problematic
<Marth64>
no not at all, i think that we leave ass alone for now
<NanananaLeader>
also, are you aware of bunch of similar subtitle filtering related patches from some guy that are iirc still on github fork?
<Marth64>
yes, developer did not address feedback. i asked them if they would continue the work last April and they said yes. but i have seen nothing so I plan to tackle it this year
<Marth64>
afk for a bit. have a good day all
NanananaLeader has quit [Quit: Client closed]
witchymary has quit [Remote host closed the connection]
witchymary has joined #ffmpeg-devel
<fflogger>
[newticket] root670: Ticket #11409 ([ffmpeg] Using -af and -vf with non-exhistant filter succeeds if input lacks required stream type) created https://trac.ffmpeg.org/ticket/11409
Everything has quit [Quit: leaving]
Kei_N_ has quit [Read error: Connection reset by peer]
Kei_N has joined #ffmpeg-devel
mkver has quit [Read error: Connection reset by peer]