Player is loading...
Keynote speech from ICC alumni David Imseng
David Imseng, Recapp
Tuesday, Aug. 27, 2019 · 3:35 p.m. · 17m 11s
ICC alumni testimonial
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
come everybody it's my pleasure to be here today um as i couldn't use them i says yeah i don't i
00:00:07
recap participate it's a t. i. c. c. in two thousand um
00:00:12
thirteen and actually want that year i participated the second time i thing where one or two years ago
00:00:18
we want again i think um but if they were focus on on recap
00:00:24
so um i come from media up so i did my p. h. d. here did yep
00:00:28
uh um i have a p. h. d. in multilingual speech recognition so my background is really technical
00:00:34
um i don't have an m. b. a. yeah i'm sorry for that
00:00:38
um so how do i see i see see in retrospect so um
00:00:44
that's how i what visualise i. c. c. five years after so you
00:00:49
are the fish x. there um this this stuff should be fish x.
00:00:54
why are you fish x. you are full of potential you have plenty of ideas in the region
00:01:00
but you are very well no but as well you're probably not
00:01:04
even found a dozen as a as a company so it's very
00:01:08
where a purdue stage and the beginning but you play games and you want to be the
00:01:14
big shark and the big fish and that's what i see is for sure you have three weeks
00:01:19
from a as much as you can and if you don't work as much as you would play a very
00:01:25
addictive game then you don't work enough to that i think is the important thing for the next three weeks um
00:01:33
sleep you can afterwards um it's very important to be they're doing the three weeks and benefit from
00:01:40
the coaches the experts um whatever you gain here knowledge um you won't lose anymore
00:01:48
um how us our i. c. c. journey beans of um in twenty thirteen
00:01:54
we pitched about meetings that's why at that time recap was called the meeting
00:02:00
um so we have this problem off of an efficient meetings everybody um i think knows that
00:02:07
the the average cost of a meeting in america is and one
00:02:11
thousand two hundred dollars and the average cost means just that if you
00:02:16
some the salary of the people that they are
00:02:20
sitting in the meeting and being not so protective
00:02:24
because one of the three meetings has been charges being unproductive which is a failure
00:02:30
um and if you do a good followup you can improve few meetings excess weight by twenty percent
00:02:36
so that has been a study that was conducted by right isn't um back then
00:02:40
i'm not sure if that changed much for me it's still the same experience if you
00:02:45
i have a scribe call 'em until everybody's there and after
00:02:48
twenty halos in tears and and nobody problems anymore you can start
00:02:53
so i think that has not changed much um at the time we had technology and i. p. so we have a
00:03:00
lot that interface that you see there's actually that all that
00:03:03
comes from from my c. c. v. really started at zero
00:03:07
um with a lot the u. i. um or you can do
00:03:10
um speech recognition speaker identification um
00:03:13
you could search your meeting afterwards um
00:03:17
so basically you could put search terms um this is
00:03:21
the different people that speak um uh they of values option
00:03:25
you can even do a kind of a topic extraction so you can visualise you meeting was attacked clout
00:03:30
in order to see what happen um yeah the
00:03:34
then submitted um buttons um the pitching and so on
00:03:39
um yeah that was the end of advice easy for our team and
00:03:43
backing twenty thirteen so we have been a team of four people um
00:03:48
by the way if you do started we lose here um that has been five years ago um
00:03:55
so it was a tough time um next to me my wife erica that
00:04:00
has been in a bank until then so she did all the financial centre budgeting and so on
00:04:06
and two friends of me um coding in these german are low
00:04:11
with american and at that time uh we built um as i say that interface you have
00:04:17
the mobile that to record meetings and so on so that was the end of i. c. c.
00:04:22
and then actually that's what happened afterwards so we won that in september um then we
00:04:27
did the i. m. b. started competition in november enabled event to pitch in in barclay
00:04:34
in california and we won the grand prize there that was a lot of fun to win against
00:04:40
us people um we have been finalists here at the local um kind
00:04:45
of competition and in september we value to the incubator of the arc
00:04:51
i which i think is still part of this or not anymore anymore so at the time it was the our crews there um
00:04:58
the founder d. the company in october um twenty fourteen so almost one
00:05:04
year or more than one year after i. c. c. has finished ah b.
00:05:10
incorporate the company because we needed to to send the first
00:05:14
invoices which was a pretty good reason to have a company
00:05:18
um we got an innovation project um funded which is very important for us at that time to get money
00:05:24
and then we started right them is our first customer which was um the parliament of monday
00:05:30
um i mean yeah now we have seen like in one year we change for
00:05:34
me doing it more efficient to parliament of wiener so how did that change happen
00:05:40
and this has to do with the after i see see she device unit that's how i see this
00:05:45
situation you have plenty of incubators you have started competition um as a site for us it was the arc um
00:05:51
i am the is there a bench because they're um see marks his next business centres are there um
00:05:58
all these supporters i'm like it's the same for group
00:06:01
meet yell i think they have two very important things
00:06:06
um because like after i see see you see red is
00:06:09
always you so you're not the egg anymore you had a fish
00:06:13
but you're a smaller fish um and there are still sharks um in in such a fish tank
00:06:20
and one thing that is important is these
00:06:23
guys whoever support you coaches advises et cetera
00:06:27
they raise like stop signs and they kind of protect you to get a bit by um big fishes
00:06:34
and the other thing that's also important in my opinion at that
00:06:37
stage is easy if you had a fish tank you have some
00:06:41
some public that watches you and you need to get some
00:06:44
traction you need to have something in the media you need to
00:06:48
make people get to know you and for this kind of
00:06:51
things um all these organisations that are out there are very helpful
00:06:56
and again it's important to get the context doing the next two weeks so whatever people you meet during
00:07:02
the next three weeks you have contact you can get back to them and then they would have you
00:07:07
i'm in getting traction and in in giving advice
00:07:12
so actually doing that time what happens to us is i'm sorry for that's like that got somehow
00:07:19
frank um well these are the meetings um these are parliament so why
00:07:25
did you change from there to here um and the reason is um
00:07:29
mainly because of the third access here which i'm going to talk about in a in a minute
00:07:34
um if you do speech recognition actually there are two important factors so one is the
00:07:39
signal quality which and usually isn't these axes and the other is the scenario complexities of
00:07:46
i'm down here is one speaker many speaker and this is good audio quality and that's bad audio quality
00:07:52
um so the meetings are the most difficult off of whatever is on that meeting
00:07:59
but difficult just means you need to have high technology and great technology you can there
00:08:05
are a lot but places such as this the yep research is you'd you'd um and
00:08:10
as you know now that are plenty of researches you that that had
00:08:13
you to solve problems but the much more important taxes is this one here
00:08:19
and it's the willingness to record meetings and to transcribe them
00:08:24
and also the need of having transcribed meeting in the sense that they are transcribed word by word
00:08:30
um and that's something that we realised very quickly when you're page two
00:08:36
to be easy for example that all those you but um my meetings are inefficient but you'll never record my meetings
00:08:43
we don't even take secretaries did important meetings or why the hell should record them and they're not allowed to the clap
00:08:49
um so there was really um no willingness to
00:08:54
to use such a tool and there was no real need of transcribing meetings
00:08:59
and at the same time we realised that parliament on the other hand by already record meetings
00:09:05
um they put them wherever you want because it's in general it's
00:09:09
it's public data where there is debate the the parliament and they need
00:09:13
row by row transcription to actually have people that
00:09:17
work on the word by word transcripts which is ah
00:09:21
well it really a difficult chop the trace down um and that's
00:09:27
what we did then is recap because recap is not re meeting anymore
00:09:31
but we kept stands for the apt to recap so how can you recap with delight quickly
00:09:37
i'm on a meeting and before um recap that's actually the work of this um
00:09:44
people that to the manor transcripts um so the rock for hours
00:09:48
and that the end all they have is that p. d. f. document
00:09:52
a p. d. f. document is put on the net and you know like if you have four hours of
00:09:57
meetings that means if you literally transcribe but whatever you
00:10:01
have about sixty pages i'm off of form of text
00:10:06
um i'm not sure how many people really read that's sixty pages
00:10:11
and then with recap what you have is the people that to the transcripts they have
00:10:16
a different job so they still need to correct the mistake of speech recognition because speech recognition
00:10:23
robot use one hundred percent correct them text but the text can already be
00:10:28
ninety or ninety five percent correct and to people than they just correcting mistakes
00:10:34
and if we do speech recognition and what we can i
00:10:38
can show them to the public is an audio visual archive bakeries
00:10:43
if we do speech recognition for each word we know exactly who shape that at what time
00:10:49
so you can very precisely do keyword searches um and that's
00:10:53
something which really puts the work that the people have anyway um
00:10:58
to the public and then you have a tool that that is used um and that
00:11:05
that can help for example um like um the courts so one
00:11:10
case that we had is that the um the parliament of always bilingual
00:11:14
to people speak either french or german and all the laws all the bilingual so they're translated and that ah
00:11:23
a southerly there was a law all where there was a contradiction in the translation so
00:11:28
there was a french version of all on the german version of rubble and they contradicted themselves
00:11:33
so what happened is that the court instead off going into p. d. f. documents and reading
00:11:39
um to to see what they discussed actually how they decided how the law should be
00:11:45
they used our interface m. search was key word for that for that
00:11:49
for the positions where read that last debated and they very quickly found the
00:11:54
the crucial some point of the bait and they were able to correct
00:11:58
the mistakes um in the to the in the i'm sorry in the translation
00:12:04
so that is um uh you a tool which is not only
00:12:08
a nice to have but it can also help the justice for example
00:12:14
so typical custom applications that's what we did so we have that will op done an interface
00:12:20
where you can um efficiently correct the mistakes of speech
00:12:24
recognition and the nice thing about the cloud based solution
00:12:29
that does speech recognition and that people correct that the the mistakes is that we always get
00:12:35
data we get annotated data um the system gets better and better it loans and so on
00:12:42
so we have that interface but people to the corrections i would have you to
00:12:47
billings by and french anyways i'm not sure how how many of you understand that
00:12:51
um says you may get the slides and you can click on the links later if that interests you
00:12:56
um but what is important is that our tax does not just
00:13:00
roll text it's formatted text so special in german be at mixed casing
00:13:06
so some words are written application some lower case and it's not the same
00:13:11
uh we do put 'em punctuation automatically
00:13:15
and um as a set the system learns um if if um if it's used
00:13:22
um then the the interactive protocol it looks like that and there
00:13:26
again you can go internet so v. s. which stands for valet
00:13:30
but we kept the th you have that search interface you can put the key work and you'll find um
00:13:37
the the important um point or the point where that key what was said in the meeting very quickly
00:13:44
um yeah and then we added some other stuff
00:13:46
like um politicians can share the debates and phase burke
00:13:51
um it's particularly the less interesting years like the one where
00:13:56
they will get reelected i'm sorry they kind of generate some some
00:14:02
um some recap from advertisements just by by sharing your thumb on face burke
00:14:09
um yeah and then at some point in time i think there is a third phase um
00:14:15
of your company which i feel we are slowly getting into the beaker is
00:14:19
if you had in in kobe to um you have a lot of support in
00:14:23
this credit goes down and now after five years so recap is now five years old
00:14:29
we need to swim ourselves so we are that fish we're all the bigger than other fishes
00:14:34
um but there's still a very small fish there's a lot of competition um of course business
00:14:39
is not always easy and you need to make sure that ah no big fish catches you
00:14:45
um and er although that all these investors so this is the best
00:14:50
um so you to make sure what you eat them when you get money from and so on
00:14:55
so that's um a couple off of customers we work with um today the
00:15:02
collaborative is the this this lady on t. v. uh we collaborate these various um
00:15:08
cantons that too um these and transcripts the the the word by word um transcriptions
00:15:14
a weaver for example this girl logically is active in
00:15:18
banking we've worked with our who's who's doing media monitoring
00:15:21
a river 'cause archives um searches me detective or the the we've got this this
00:15:27
come um and there are a couple of other um customers as well um in german
00:15:34
some kind of statements about the cap which is a kind
00:15:37
of a one minute pitch so today um this recap you can
00:15:41
and transcribe or you can generate protocols of fifty percent more efficient
00:15:46
and that's what they should add various places and forty minutes after for
00:15:50
our debate you half the first version of to automatically generate the protocol
00:15:56
so in the same evening even you already have the
00:15:59
videos online of the political debate that can be searched
00:16:04
and these are the cab you would rate them a week a
00:16:07
recap also allows you to to index archive movies this one eight
00:16:13
of the time so usually for one hour video you need either
00:16:16
of all i got was of work um by using recap you can
00:16:21
um have one hour of work for one of the the uh and in in media monitoring
00:16:27
as i said before so each twenty four hour should be transcribe more than two hundred hours
00:16:33
in french german and swiss german dialect is our and speech technology
00:16:39
and today really cab or or eight people um so my wife is
00:16:44
still there um then we have the c. d. uh we have um
00:16:48
several software engineers you have a linguist and we also have some interns
00:16:54
we are right now um based interface we still have an office in martini and we have an office and burn
00:17:01
um yeah that's it um thank you for your attention and if
Recommended talks
Présentation des 50 ans de Visarte Valais-Wallis
Elisabeth Fux Mattig et Helga Zumstein, Présidente et membre du Comité de Visarte Wallis
Oct. 4, 2023 · 10:31 a.m.
Die Schweizerische Post – innovative Taktgeberin auf der letzten Meile
Yves König, Leiter Smart Urban Logistics, Die Post
March 21, 2019 · 10:46 a.m.
868 views