►
From YouTube: TPAC 2019 WebRTC meeting @Friday part 3
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
D
Street
and
transceiver-
and
we
now
see
us
it's
a
bit
backward
to
have
a
street
comedian
s
who
she
stands
as
human
right
and
relationships.
Peggy's
we
remove
track
tracks,
don't
exist,
I
propose
we
remove
this
track
as
well
proposal
is
that
we
there
is
still
the
notion
of
you
might
enter
fires.
It's
part
of
the
sender
and
receiver.
I
have
a
slot
extreme,
swap
so
proposal
a
is
removed
the
medium
stature,
but
keep
stream
identifiers
on
sender
and
receivers.
That
proposal,
be
yes,
say
well,
okay,
is
removing
anything
in.
E
D
They
need
to
be
the
same
cross,
get
that's
call,
so
you
can
you
get
the
same
sense
of
it
and
because
it's
specified,
we've
been
implementations
have
been
just
coming
up
with
strings
like
RTC
new
using
track
understand
your
understa
one
problem.
Yes,
you
can
sort
of
predict,
predict
the
IDS
even
or
you
go
guess
s,
which
means
that
some
applications
might
start
relying
on
implementation,
specific
behavior
or
like
if
I,
deep,
suddenly
changes.
D
Crisis
so
that
I
propose
that
we
require
I
needs
to
be
unpredictable
in
exact
wording
of
the
parent
can
be
discussing
the
PR,
but
I
want
to
propose
that
we
we
make
it
basically
random,
looking
and
yeah
right
what
suggestion
hard,
let's
use,
base64
a
shop
or
the
predicting
will
string.
I
still
think
you
need
to
seed
it
somehow
so
that
it's
different
across
sessions.
D
E
D
D
It's
like
actually
that's
where
they're
running
negative,
I'm
sure
everything
this
current
thing.
We
have
no
weight
as
we
don't
expose
the
encoding,
parameter,
stats
or
recurring
parameters,
and
so
the
proposal
here
is
that
we
expose
the
same
information
as
is
available
in
the
parameters.
Api,
your
skippers
line,
I.
D
Days
and
I
know
your
previous
lectures
about
current
direction.
Your
might
have
objections
of
that
this
thing,
but
so
the
thing
we
must
we
really
want
to
have
as
it's
just
right
now.
You
have
multiple
a
planar
key
piece.
That's
up
next
one
per
single
cast
layer
and
you
don't
know
which
one
is
which
so
you'd
have
to
just
guess,
based
on
resolution
or
something
that's
not
pretty
reliable.
So
we
definitely
want
to
wait
and
random
either
we
do.
This
slide
is,
as
we.
F
D
E
E
G
D
D
A
E
A
D
D
A
I
think
none
of
my,
like
none
of
my
colleagues,
are
on
the
on
the
call,
because
too
early
in
Finland,
so
I
would
definitely
take
it
to
the
list
for
if
they
want
any
of
the
additional
things,
I
think
I
feel
get.
If
get
parameters
exists,
it
should
not
be
a
problem
for
us
to
just
go
with
the
single
red
line
and
they
are
bounced
at
I.
A
Believe
we
had
some
discussions
in
PR,
482
or
in
issue
third
395,
which
made
us
fill
these
two
out
these
two
proposals
out
for
82
and
for
85,
so
they
don't
have
a
strong
objection
or
counterproposal
right
now.
I'm,
fine
with
dot
red
but
I
would
want
the
team
to
people
on
our
side
to
say
that
it's
okay
for
them.
E
Twice
in
a
row
you
can
just
in
transaction
ID,
which
is
not
correct,
because
it
assumes
that
you
only
with
you
Hoggett
progress
is
because
you
intend
to
set
progress
right.
But
if
you
have
library
that
we
call
some
function,
then
that
library
now
bother
me
other
reasons
to
get
the
parameters
it'll.
Now,
when
I
get
back
to
the
main
routine
I've
changed
the
transaction
ID,
which
means.
E
H
D
A
I
Yes,
either
rather
I
have
a
comment
here
which
is
so
along
with
SVC.
We
on
this
new
weird
thing,
which
is
sama,
cast
without
separate
a
separate
entity
streams.
Well
that
separate
our
deepest
needs
right.
That's
the
s
modes
and
WebRTC
ESPC.
So
when
you
talk
about
layer
sheet,
I
think
you
should
talk
about
how
you
identify
that,
like
typically
what
you
think
of
as
a
tid
and
an
S
ID.
I
D
I
I
It's
basically
spatial
at
the
idea.
Nice
I
didn't
want
numbers
yeah,
so
it's
spatial
I
can't
in
temporal
ID.
But
the
thing
to
realize
is
that
it's
possible
to
have
a
it's
possible
to
have
a
spatial
ID
represent
a
simulcast,
including
within
a
month.
It
can
be
multiple
simulcast
encodings
within
single
our
key
stream.
That's
town
code
such
as
81.
I
So
let's
an
example:
you
have
an
a
you
know:
si
B
3
could
could
mean
you
had
3
spinkle,
SVC
layers
or
depending
on
the
mode.
It
couldn't
be
simulcast
layers,
also
within
a
single
RTD
stream.
Anyway,
I'm
basically
saying
I,
think
I
think
you
should
describe
with
the
layer
of
exactly
a
little
bit
better
in
those
little
areas
and
maybe
a
moment
delineation.
So
your
clear
way
yeah
so.
A
I
am
actually
in
I
think
we
probably
cannot
be
I
think
we
should
probably
discuss
this
in
a
separate
meeting,
because
it
quality
sounds
more
complicated
because
my
question
next
is.
We
then
need
a
dependency
index
to
see
what
is
dependent
on
what,
because
how
like
what,
if
all
simulcasted
within
one
RTP
how's
that
well.
I
D
Know
knowing
the
issue
about
how
do
we,
how
do
we
identify
the
layer
or
how
do
we
know
the
relationship
layers?
This
is
the
basic
idea
of
having
one
layer,
10cc
layers,
that's
objects
per
layer
with
an
individual
with
height
sense.
My
sense
does
that
make
sense,
can
we
yeah
I
mean
I
I?
Think
what
you're
trying
to
do.
I
Where
you
might
want
to
know,
as
an
example
would
be
well
there's
you
know
the
overall
loss
I'm
experiencing
on
the
whole
thing,
and
certainly
all
the
layers
to
mine
right.
That's
a
useful
thing
to
know
that
that
might
well
be
sufficient
in
the
sense
that
you
know
it's
possibly
a
difference
in
the
loss
that
you
get
between
layers
as
to
whether
that's
meaningful
to
any
congestion,
control
of
them
and
I.
Don't
know
why
it
would
be.
I
You
would
just
look
kind
of
look
at
your
overall
bandwidth
availability
and
packet
loss,
and
me,
you
know
the
big
question
I
have
is:
what
are
you
trying
to
I
mean
you?
Might
you
might
basically
look
at
this
in
retrospect
and
try
to
conclude,
you
know
what
your
strategy
ought
to
have
been
with
respect
to
dropping
layers
are
doing
something
you
might
decide.
Hey
I
left
this
finance
the
way
I'm
going
to
along
and
was
just
experiencing
a
lot
of
loss
once
I
dropped.
It
things
got
there.
B
E
D
D
Detecting
that
you
go
okay,
so
it's
the
current
current
metrics
it's
hard
to
distinguish
between
inches
and
level
playing
rates
can
bowling
against
s.
You
can
calculate
averages,
but
you
know
unless
you're
pulling
way
more
often
than
you
should
be,
it's
hard
to
know,
and
so
you
know,
one
proposal
was
to
just
say:
they're
a
my
pitch
is
such-and-such
within
milliseconds,
but
this
is
a
problem
like
to
200
milliseconds
between
two
consecutive
frames.
That's
great!
If
you're
happy
here
running
it
up
like
that.
Yes,
right
video,
but
you're
running
a
penny.
D
I
visited
years,
it's
actually
in
the
edges
table,
but
it's
hard
to
define
what
a
gauge
is.
So
the
proposal
a
is
to
just
take
the
maximum
interframe
delay
that
was
recorded
during
the
last
second,
which
means
yeah.
So
if
you
call
guess
that's
one
leadership
once
per
second
you'd
have
a
very
good
idea,
but
it's
just
and
this
will
work.
Even
if
you
have
my
cup,
you
only
get
one
frame
per
10
seconds.
F
D
Zero
point
the
day
will
be
10,
I
guess
and
it's
it's
easy
to
understand,
but
it's
a
maximum
and
it
sort
of
has
assumptions
about
how
often
you're
pulling
guess
next,
if
you're
pulling
it
to
once
every
10
seconds
you
miss
Osen
game.
So
that's
a
argues
against
proposal.
A
argument
for
proposal
a
it's
easy
to
understand.
D
Registration
and
I
haven't
run
the
exact
wording
here,
but
we
defined
a
glitch
as
a
percentage
of
deviation
from
the
average
frame
rates.
That's
based
on
recent
frames
or
recent
times.
Okay
can
I
get
a
sense
from
the
room
if
I
should
go
in
direction,
A
or
B,
or
something
else.
I'd
like
the
P
are
trying
to
retrieve
Mitch's.
A
A
What
is
the
duration
between
in
frame
and
if
it
was
skipped
or
or
with
an
expectation
if
I
am
expecting
500
may,
like
I'm,
expecting
200,
milliseconds
and
I
get
something
at
250?
Doesn't
it
already
know
that
I
was
expecting
something
at
200
milliseconds
apart
and
define
it
on
per
frame
basis?
The
deviation
because
I
don't
like
the
first
one,
because
it
forces
me
to
pull
and
frequently
and
if
I
don't
want
to
pull
frequently
I
won't
get
this
thing.
I
won't
get
a
heuristic
over
time.
A
D
H
H
H
B
G
H
Quality
attack,
you
know
we
have
babies
to
shoot
mister.
Definitely
you
have
the
detail
defer
and
there
is
the
red
ring
buffer.
Then
there
is
going
to
be
Kampala
Beach
from
inertia
and
on
eventually
they
will
be
buffer
before
and
after
that
will
contribute
to
the
perception
of
the
beach.
We
understand.
Of
course,
you.
H
D
J
E
A
H
So,
let's
make
some
assumptions
here,
even
in
the
XF
UK's
or
the
SEC
Jason
240fps
is
supposed
to
be
stable
by
segments.
So
it's
not
the
variation
allotted
fast
way
to
chill.
There
will
be
a
step
at
one
time.
Then
it
will
just
a
blonde
is
again
not
something
we
can.
We
can
be
born
and
the
stability
of
the
end
yes
or
the
either
from
the
limit
is
the
same.
That.
D
H
To
detect
the
expectation,
but
we
can
infer
these
petitions
supposing
that
the
ambience
is
constant,
always
for
certain
amount
of
time
in
my
drop-
and
you
know
have
one
one
friend
to
the
next
level
in
remains
deadlock,
so
the
curve
recently
supposedly
a
linear
but
still
constant,
burn
time
segment.
I.
B
C
D
What
can
we
report
if
we
have,
if
we're
both
CPU
limited
and
bandwidth
limited
at
the
same
time,
and
the
reason
why
this
hard
to
answer
is
because
these
are
not
independent
variables
and
your
are
your
kind
of
guesstimate
beliefs,
you
you
you,
you
try
to
do
something,
and
you
notice
that,
although
I'm
using
a
lot
of
CPU,
so
you
go
down,
this
will
affect
the
bandwidth
as
many
badness
can
the
same.
You
start
increasing
it,
and
so.
H
C
D
B
B
A
Sometimes
people
have
other
things
running
on
their
CPU
on
their
on
their
systems,
which
causes
the
CPU
search.
The
CPU
tends
to
be
quite
interesting
when
it
happens
and
for
bandwidth
there
are
other
ways
to
like
make
some
assumptions.
Its
bandwidth
is
actually
the
case,
because
you
can
do
some
pre-call
test.
You
can
like
measure
what's
happening
in
during
the
car.
You
can
do
it
after
the
call.
You
can
do
it
in
background
so
see
in
the
case
of
bandwidth.
We
have
a
lot
of
information
and
CPU.
A
D
E
E
K
E
D
A
F
D
All
these
samples,
okay,
the
problem
is,
we
have
we
have
these
audio
energy
capital
cycle
duration,
sample
counters.
We
have
various
eco
eco
returned
lost,
even
there's
various
encounters
and
metrics
based
on
and
audio
samples,
and
all
these
as
well
is
good,
except
we
didn't
really
think
about
channels
right.
D
If
you
have
multiple
channels,
you
have
less,
you
have
we
can
have
stereo
or
whatever,
and
it's
it's
not
kidding
what
you
should
do
with
these
counters
right,
if
I
have
twice
as
many
channels
to
ensure
these
counters
go
out
twice
as
fast
or
not
current
implementation.
This
is
not
like,
even
though
you
have
two
channels
both
receiving
ten
samples
and
the
county
does
not
go
up
by
20,
because
I
by
10
do
we
have
any
case,
looks.
D
D
D
Having
a
simple
version
of
the
minute,
thanks
give
it
or
the
proposed
solution:
yes,
okay,
good
moving
on
and
right,
if
you
do
an
iced
restart,
you
can
get
the
same
ice
canvas
at
night,
nice
candidate,
pairs,
that's
objects,
and
you
don't
know
which
you
know
from
which
is
generation.
These
belong
to
a
proposal.
Aim
is
to
and
I
see
you
brag
to
Garcia
nice
transport,
and
that
way
you
can
tell
a
proposal
be
makes
a
proposal
a
would
be
consistent
with
trend
implementation.
D
You
just
add
another
cue
proposal,
be
it
is
to
say
that
in
print
implementation
is
a
bag
like
if
you
do
an
ice
cream
start,
even
if,
if
the
candidate
is
the
same
ip
address,
you
should
say
this.
This
doesn't
conceptually
would
refer
to
the
same
object
anymore,
which
means
div
should
be
different.
So
it's
a
bogman
that
you
have
the
same
ID
before
and
after
and
then
way.
The
way
you
tell
that
they
don't
belong
to
the
same
ice
generation
is
that
they
were
notes.
D
A
Anywhere
that
would
depend
on
make.
It
also
depends
on
make
before
breakfast
is
make
before
break
as
in.
If
you
already
have
a
connection
into
an
ice,
we
start
it
doesn't
reset.
Your
existing
candidates,
like
you,
can
still
use
your
like.
You
are
still
using
your
old
transport
until
the
ice
restart
is
complete.
E
B
D
Creations
of
labour
channels,
and
then
the
first
one
is
well
there's.
No
there's
no
HTTP
transferred
stats,
and
the
question
here
is:
should
we
exposed
runs
of
time
measurements
based
on
the
setp
passport,
because
currently
we
have
run
at
times
based
on
our
TCP
and
bring
some
ice
things
I
heard
of
the
Firefox
does
not
do
ice
things
as
some
comments
and
they
she
was
saying
that,
so
it
would
be
useful
to
have
a
round-trip
time,
at
least
on
the
SCTP.
D
Yeah,
the
Impressionist
has
been
assess
the
current
smooth,
the
round-trip
time
in
seconds,
which
sounds
very
under
specified
to
me.
So
I
don't
know
if
that
might
be
a
question
for
the
PR
of
the
question
in
this
group
is:
should
we
add
one
SC
on
each
piece?
Part
STP
transports.
That's
object
with
whatever
round-shaped
time
we
find
based
on
SMTP
RFC's.