topic_shift
bool 2
classes | utterance
stringlengths 1
7.9k
| session_id
stringlengths 7
14
|
|---|---|---|
false |
OK.
|
QMSum_78
|
false |
We 're running a little short here.
|
QMSum_78
|
false |
That 's fine.
|
QMSum_78
|
false |
We , uh , uh , cou trying to
|
QMSum_78
|
false |
I 'm finished.
|
QMSum_78
|
false |
eh , was p Well , you know , the thing I 'm concerned about is we wanted to do these digits
|
QMSum_78
|
false |
Oh , yeah.
|
QMSum_78
|
false |
and and I haven't heard , uh , from Jose yet.
|
QMSum_78
|
false |
Oh , yes.
|
QMSum_78
|
false |
OK. What do you want ?
|
QMSum_78
|
false |
Mm - hmm.
|
QMSum_78
|
false |
So
|
QMSum_78
|
false |
We could skip the digits.
|
QMSum_78
|
false |
Uh
|
QMSum_78
|
false |
We don't have to read digits each time.
|
QMSum_78
|
false |
Uh I I I think it you know , another another bunch of digits. More data is good.
|
QMSum_78
|
false |
OK.
|
QMSum_78
|
false |
Yeah. Sure.
|
QMSum_78
|
false |
So so I 'd like to do that. But I think , do you , maybe , eh ? Did you prepare some whole thing you wanted us just to see ?
|
QMSum_78
|
false |
Yeah. It 's it 's prepared.
|
QMSum_78
|
false |
Or what was that ? Yeah.
|
QMSum_78
|
false |
Oh , k Sorry.
|
QMSum_78
|
false |
Uh , how long a ?
|
QMSum_78
|
false |
I I think it 's it 's fast , because , uh , I have the results , eh , of the study of different energy without the law length. Eh , um , eh , in the in the measurement , uh , the average , uh , dividing by the by the , um , variance. Um , I th i
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
the other , uh the the last w uh , meeting eh , I don't know if you remain we have problem to with the with with the parameter with the representations of parameter , because the the valleys and the peaks in the signal , eh , look like , eh , it doesn't follow to the to the energy in the signal.
|
QMSum_78
|
false |
Yes. Right.
|
QMSum_78
|
false |
And it was a problem , uh , with the scale.
|
QMSum_78
|
false |
With what ?
|
QMSum_78
|
false |
Eh , the scale.
|
QMSum_78
|
false |
Scale.
|
QMSum_78
|
false |
Scale.
|
QMSum_78
|
false |
Eh , and I I change the scale and we can see the the variance.
|
QMSum_78
|
false |
OK. But the bottom line is it 's still not , uh , separating out very well.
|
QMSum_78
|
false |
Yeah. Yeah.
|
QMSum_78
|
false |
Right ?
|
QMSum_78
|
false |
The distribution the distribution is is similar.
|
QMSum_78
|
false |
OK. So that 's that 's that 's enough then. OK.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
No , I mean , that there 's no point in going through all of that if that 's the bottom line , really.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
Mm - hmm.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
So , I I think we have to start Uh , I mean , there there 's two suggestions , really , which is , uh what we said before is that ,
|
QMSum_78
|
false |
Mmm , yeah.
|
QMSum_78
|
false |
um , it looks like , at least that you haven't found an obvious way to normalize so that the energy is anything like a reliable , uh , indicator of the overlap.
|
QMSum_78
|
false |
Yeah. Yeah.
|
QMSum_78
|
false |
Um , I I 'm I 'm still a little f think that 's a little funny. These things l @ @ seems like there should be ,
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
but but you don't want to keep , uh keep knocking at it if it 's if you 're not getting any any result with that. But , I mean , the other things that we talked about is , uh , pitch - related things and harmonicity - related things ,
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
so which we thought also should be some kind of a reasonable indicator. Um But , uh , a completely different tack on it wou is the one that was suggested , uh , by your colleagues in Spain ,
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
which is to say , don't worry so much about the , uh , features.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
That is to say , use , you know , as as you 're doing with the speech , uh , nonspeech , use some very general features.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
And , uh , then , uh , look at it more from the aspect of modeling.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
You know , have a have a couple Markov models and and , uh , try to indi try to determine , you know , w when is th when are you in an overlap , when are you not in an overlap.
|
QMSum_78
|
false |
Hmm.
|
QMSum_78
|
false |
And let the , uh , uh , statistical system determine what 's the right way to look at the data.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
I I , um , I think it would be interesting to find individual features and put them together. I think that you 'd end up with a better system overall.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
But given the limitation in time and given the fact that Javier 's system already exists doing this sort of thing ,
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
uh , but , uh , its main limitation is that , again , it 's only looking at silences which would
|
QMSum_78
|
false |
Yeah. Yeah.
|
QMSum_78
|
false |
maybe that 's a better place to go.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
Mm - hmm.
|
QMSum_78
|
false |
So.
|
QMSum_78
|
false |
I I I think that , eh , the possibility , eh , can be that , eh , Thilo , eh , working , eh , with a new class , not only , eh , nonspeech and speech , but , eh , in in in the speech class ,
|
QMSum_78
|
false |
Mm - hmm. Mm - hmm.
|
QMSum_78
|
false |
dividing , eh , speech , eh , of from a speaker and overlapping , to try to to do , eh , eh , a fast a fast , eh , experiment to to prove that , nnn , this fea eh , general feature , eh , can solve the the the problem ,
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
and wh what nnn , how far is
|
QMSum_78
|
false |
Maybe. Yeah.
|
QMSum_78
|
false |
And , I I have prepared the the pitch tracker now.
|
QMSum_78
|
false |
Mm - hmm.
|
QMSum_78
|
false |
And I hope the the next week I will have , eh , some results and we we will show we will see , eh , the the parameter the pitch , eh , tracking in with the program.
|
QMSum_78
|
false |
I see.
|
QMSum_78
|
false |
And , nnn , nnn
|
QMSum_78
|
false |
Ha - h have you ever looked at the , uh , uh Javier 's , uh , speech segmenter ?
|
QMSum_78
|
false |
No. No.
|
QMSum_78
|
false |
No.
|
QMSum_78
|
false |
Oh. Maybe m you could , you kn uh show Thilo that.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
Yeah. Sure.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
Cuz again the idea is there the limitation there again was that he was he was only using it to look at silence as a as a as a as a p putative split point between speakers.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
But if you included , uh , broadened classes then in principle maybe you can cover the overlap cases.
|
QMSum_78
|
false |
OK.
|
QMSum_78
|
false |
Yeah.
|
QMSum_78
|
false |
Yeah , but I 'm not too sure if if we can really represent overlap with with the s detector I I I used up to now ,
|
QMSum_78
|
false |
Mmm , yeah.
|
QMSum_78
|
false |
Uh
|
QMSum_78
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.