Datasets:
elplaguister
/
DTS_line_datasets

Dataset card Data Studio Files Community
topic_shift
bool
2 classes
utterance
stringlengths
1
7.9k
session_id
stringlengths
7
14
false
It seems like whatever they choose they shouldn't , you know , purposefully brain - damage a part of the system to make a worse baseline , or
QMSum_120
false
Well , I think people just had
QMSum_120
false
You know ?
QMSum_120
false
it wasn't that they purposely brain - damaged it. I think people hadn't really thought through about the , uh the VAD issue.
QMSum_120
false
Mmm.
QMSum_120
false
Mm - hmm.
QMSum_120
false
And and then when the the the proposals actually came in and half of them had V A Ds and half of them didn't , and the half that did did well and the half that didn't did poorly.
QMSum_120
false
Mm - hmm.
QMSum_120
false
So it 's
QMSum_120
false
Mm - hmm. Um.
QMSum_120
false
Uh.
QMSum_120
false
Yeah. So we 'll see what happen with this. And Yeah. So what happened since , um , last week is well , from OGI , these experiments on putting VAD on the baseline. And these experiments also are using , uh , some kind of noise compensation , so spectral subtraction , and putting on - line normalization , um , just after this. So I think spectral subtraction , LDA filtering , and on - line normalization , so which is similar to the pro proposal - one , but with spectral subtraction in addition , and it seems that on - line normalization doesn't help further when you have spectral subtraction.
QMSum_120
false
Is this related to the issue that you brought up a couple of meetings ago with the the musical tones
QMSum_120
false
and ?
QMSum_120
false
I have no idea , because the issue I brought up was with a very simple spectral subtraction approach ,
QMSum_120
false
Mmm.
QMSum_120
false
and the one that they use at OGI is one from from the proposed the the the Aurora prop uh , proposals , which might be much better. So , yeah. I asked Sunil for more information about that , but , uh , I don't know yet. Um. And what 's happened here is that we so we have this kind of new , um , reference system which use a nice a a clean downsampling - upsampling , which use a new filter that 's much shorter and which also cuts the frequency below sixty - four hertz ,
QMSum_120
false
Right.
QMSum_120
false
which was not done on our first proposal.
QMSum_120
false
When you say " we have that " , does Sunil have it now , too ,
QMSum_120
false
I No.
QMSum_120
false
or ?
QMSum_120
false
No.
QMSum_120
false
OK.
QMSum_120
false
Because we 're still testing. So we have the result for , uh , just the features
QMSum_120
false
OK.
QMSum_120
false
and we are currently testing with putting the neural network in the KLT. Um , it seems to improve on the well - matched case , um , but it 's a little bit worse on the mismatch and highly - mismatched I mean when we put the neural network. And with the current weighting I think it 's sh it will be better because the well - matched case is better. Mmm.
QMSum_120
false
But how much worse since the weighting might change how how much worse is it on the other conditions , when you say it 's a little worse ?
QMSum_120
false
It 's like , uh , fff , fff um , ten percent relative. Yeah.
QMSum_120
false
OK. Um.
QMSum_120
false
Mm - hmm.
QMSum_120
false
But it has the , uh the latencies are much shorter. That 's
QMSum_120
false
Uh - y w when I say it 's worse , it 's not it 's when I I uh , compare proposal - two to proposal - one , so , r uh , y putting neural network compared to n not having any neural network. I mean , this new system is is is better ,
QMSum_120
false
Uh - huh.
QMSum_120
false
because it has um , this sixty - four hertz cut - off , uh , clean downsampling , and , um what else ? Uh , yeah , a good VAD. We put the good VAD. So. Yeah , I don't know. I I j uh , uh pr
QMSum_120
false
But the latencies but you 've got the latency shorter now.
QMSum_120
false
Latency is short is Yeah.
QMSum_120
false
Yeah.
QMSum_120
false
Isn't it
QMSum_120
false
And so
QMSum_120
false
So it 's better than the system that we had before.
QMSum_120
false
Yeah. Mainly because of the sixty - four hertz and the good VAD.
QMSum_120
false
OK.
QMSum_120
false
And then I took this system and , mmm , w uh , I p we put the old filters also. So we have this good system , with good VAD , with the short filter and with the long filter , and , um , with the short filter it 's not worse. So well , is it
QMSum_120
false
OK.
QMSum_120
false
it 's in
QMSum_120
false
So that 's that 's all fine.
QMSum_120
false
Yes. Uh
QMSum_120
false
But what you 're saying is that when you do these So let me try to understand. When when you do these same improvements to proposal - one ,
QMSum_120
false
Mm - hmm.
QMSum_120
false
that , uh , on the i things are somewhat better , uh , in proposal - two for the well - matched case and somewhat worse for the other two cases.
QMSum_120
false
Yeah.
QMSum_120
false
So does , uh when you say , uh So The th now that these other things are in there , is it the case maybe that the additions of proposal - two over proposal - one are less im important ?
QMSum_120
false
Yeah. Probably , yeah.
QMSum_120
false
I get it.
QMSum_120
false
Um So , yeah. Uh. Yeah , but it 's a good thing anyway to have shorter delay. Then we tried , um , to do something like proposal - two but having , um , e using also MSG features. So there is this KLT part , which use just the standard features ,
QMSum_120
false
Mm - hmm. Right.
QMSum_120
false
and then two neura two neural networks.
QMSum_120
false
Mm - hmm.
QMSum_120
false
Mmm , and it doesn't seem to help. Um , however , we just have one result , which is the Italian mismatch , so. Uh. We have to wait for that to fill the whole table , but
QMSum_120
true
OK. There was a start of some effort on something related to voicing or something. Is that ?
QMSum_120
false
Yeah. Um , yeah. So basically we try to , uh , find good features that could be used for voicing detection , uh , but it 's still , uh on the , um t
QMSum_120
false
Oh , well , I have the picture.
QMSum_120
false
we w basically we are still playing with Matlab to to look at at what happened ,
QMSum_120
false
What sorts of
QMSum_120
false
Yeah.
QMSum_120
false
and
QMSum_120
false
what sorts of features are you looking at ?
QMSum_120
false
We have some
QMSum_120
false
So we would be looking at , um , the variance of the spectrum of the excitation ,
QMSum_120
false
uh , um , this , this , and this.
QMSum_120
false
something like this , which is should be high for voiced sounds. Uh , we
QMSum_120
false
Wait a minute. I what does that mean ? The variance of the spectrum of excitation.
QMSum_120
false
Yeah. So the So basically the spectrum of the excitation for a purely periodic sig signal shou sh
QMSum_120
false
OK. Yeah , w what yo what you 're calling the excitation , as I recall , is you 're subtracting the the , um the mel mel mel filter , uh , spectrum from the FFT spectrum.
QMSum_120
false
e That 's right. Yeah. So
QMSum_120
false
Right.
QMSum_120
false
Yeah.
QMSum_120
false
Mm - hmm.
QMSum_120
false
So we have the mel f filter bank , we have the FFT , so we just
QMSum_120
false
So it 's it 's not really an excitation ,
QMSum_120
false
No.
QMSum_120
false
but it 's something that hopefully tells you something about the excitation.
QMSum_120
false
Yeah , that 's right.
QMSum_120
false
Yeah , yeah.
QMSum_120
false
Um Yeah.
QMSum_120
false
We have here some histogram ,
QMSum_120
false
E yeah ,
QMSum_120
false
but they have a lot of overlap.
QMSum_120
false
but it 's it 's still Yeah. So , well , for unvoiced portion we have something tha that has a mean around O point three , and for voiced portion the mean is O point fifty - nine. But the variance seem quite high.
QMSum_120
false
How do you know ?
QMSum_120
false
So Mmm.
QMSum_120
false
How did you get your voiced and unvoiced truth data ?
QMSum_120
false
We used , uh , TIMIT and we used canonical mappings between the phones
QMSum_120
false
Yeah. We , uh , use TIMIT on this ,
QMSum_120
false
and
QMSum_120
false
for
QMSum_120
false
th Yeah.
QMSum_120
false
But if we look at it in one sentence , it apparently it 's good , I think.
QMSum_120
false
Yeah , but Yeah. Uh , so it 's noisy TIMIT. That 's right. Yeah.
QMSum_120