elplaguister/DTS_pair_datasets · Datasets at Hugging Face

prev1 stringlengths 1 7.9k	prev2 stringlengths 1 7.9k ⌀	next1 stringlengths 1 7.9k	next2 stringlengths 1 7.9k ⌀	is_boundary bool 2 classes	session_id stringlengths 7 14
So it 's i All the noises are from the TI - digits ,	Yeah.	right ? So you i	Um They k uh	false	QMSum_194
Yeah.	right ? So you i	Um They k uh	Well , it it 's like the high mismatch of the SpeechDat - Car after cleaning up , maybe having more noise than the the training set of TIMIT after clean s after you do the noise clean - up.	false	QMSum_194
right ? So you i	Um They k uh	Well , it it 's like the high mismatch of the SpeechDat - Car after cleaning up , maybe having more noise than the the training set of TIMIT after clean s after you do the noise clean - up.	Mmm.	false	QMSum_194
Um They k uh	Well , it it 's like the high mismatch of the SpeechDat - Car after cleaning up , maybe having more noise than the the training set of TIMIT after clean s after you do the noise clean - up.	Mmm.	I mean , earlier you never had any compensation , you just trained it straight away.	false	QMSum_194
Well , it it 's like the high mismatch of the SpeechDat - Car after cleaning up , maybe having more noise than the the training set of TIMIT after clean s after you do the noise clean - up.	Mmm.	I mean , earlier you never had any compensation , you just trained it straight away.	Mm - hmm.	false	QMSum_194
Mmm.	I mean , earlier you never had any compensation , you just trained it straight away.	Mm - hmm.	So it had like all these different conditions of S N Rs , actually in their training set of neural net.	false	QMSum_194
I mean , earlier you never had any compensation , you just trained it straight away.	Mm - hmm.	So it had like all these different conditions of S N Rs , actually in their training set of neural net.	Mm - hmm. Mm - hmm.	false	QMSum_194
Mm - hmm.	So it had like all these different conditions of S N Rs , actually in their training set of neural net.	Mm - hmm. Mm - hmm.	But after cleaning up you have now a different set of S N Rs , right ?	false	QMSum_194
So it had like all these different conditions of S N Rs , actually in their training set of neural net.	Mm - hmm. Mm - hmm.	But after cleaning up you have now a different set of S N Rs , right ?	Yeah.	false	QMSum_194
Mm - hmm. Mm - hmm.	But after cleaning up you have now a different set of S N Rs , right ?	Yeah.	For the training of the neural net.	false	QMSum_194
But after cleaning up you have now a different set of S N Rs , right ?	Yeah.	For the training of the neural net.	Mm - hmm.	false	QMSum_194
Yeah.	For the training of the neural net.	Mm - hmm.	And is it something to do with the mismatch that that 's created after the cleaning up , like the high mismatch	false	QMSum_194
For the training of the neural net.	Mm - hmm.	And is it something to do with the mismatch that that 's created after the cleaning up , like the high mismatch	You mean the the most noisy occurrences on SpeechDat - Car might be a lot more noisy than	false	QMSum_194
Mm - hmm.	And is it something to do with the mismatch that that 's created after the cleaning up , like the high mismatch	You mean the the most noisy occurrences on SpeechDat - Car might be a lot more noisy than	Mm - hmm. Of that I mean , the SNR after the noise compensation of the SpeechDat - Car.	false	QMSum_194
And is it something to do with the mismatch that that 's created after the cleaning up , like the high mismatch	You mean the the most noisy occurrences on SpeechDat - Car might be a lot more noisy than	Mm - hmm. Of that I mean , the SNR after the noise compensation of the SpeechDat - Car.	Oh , so Right. So the training the the neural net is being trained with noise compensated stuff.	false	QMSum_194
You mean the the most noisy occurrences on SpeechDat - Car might be a lot more noisy than	Mm - hmm. Of that I mean , the SNR after the noise compensation of the SpeechDat - Car.	Oh , so Right. So the training the the neural net is being trained with noise compensated stuff.	Maybe.	false	QMSum_194
Mm - hmm. Of that I mean , the SNR after the noise compensation of the SpeechDat - Car.	Oh , so Right. So the training the the neural net is being trained with noise compensated stuff.	Maybe.	Yeah.	false	QMSum_194
Oh , so Right. So the training the the neural net is being trained with noise compensated stuff.	Maybe.	Yeah.	Yeah , yeah.	false	QMSum_194
Maybe.	Yeah.	Yeah , yeah.	Which makes sense ,	false	QMSum_194
Yeah.	Yeah , yeah.	Which makes sense ,	Yeah.	false	QMSum_194
Yeah , yeah.	Which makes sense ,	Yeah.	but , uh , you 're saying Yeah , the noisier ones are still going to be , even after our noise compensation , are still gonna be pretty noisy.	false	QMSum_194
Which makes sense ,	Yeah.	but , uh , you 're saying Yeah , the noisier ones are still going to be , even after our noise compensation , are still gonna be pretty noisy.	Yeah.	false	QMSum_194
Yeah.	but , uh , you 're saying Yeah , the noisier ones are still going to be , even after our noise compensation , are still gonna be pretty noisy.	Yeah.	Mm - hmm.	false	QMSum_194
but , uh , you 're saying Yeah , the noisier ones are still going to be , even after our noise compensation , are still gonna be pretty noisy.	Yeah.	Mm - hmm.	Yeah , so now the after - noise compensation the neural net is seeing a different set of S N Rs than that was originally there in the training set. Of TIMIT. Because in the TIMIT it was zero to some clean.	false	QMSum_194
Yeah.	Mm - hmm.	Yeah , so now the after - noise compensation the neural net is seeing a different set of S N Rs than that was originally there in the training set. Of TIMIT. Because in the TIMIT it was zero to some clean.	Right. Yes.	false	QMSum_194
Mm - hmm.	Yeah , so now the after - noise compensation the neural net is seeing a different set of S N Rs than that was originally there in the training set. Of TIMIT. Because in the TIMIT it was zero to some clean.	Right. Yes.	So the net saw all the SNR @ @ conditions.	false	QMSum_194
Yeah , so now the after - noise compensation the neural net is seeing a different set of S N Rs than that was originally there in the training set. Of TIMIT. Because in the TIMIT it was zero to some clean.	Right. Yes.	So the net saw all the SNR @ @ conditions.	Right.	false	QMSum_194
Right. Yes.	So the net saw all the SNR @ @ conditions.	Right.	Now after cleaning up it 's a different set of SNR.	false	QMSum_194
So the net saw all the SNR @ @ conditions.	Right.	Now after cleaning up it 's a different set of SNR.	Right.	false	QMSum_194
Right.	Now after cleaning up it 's a different set of SNR.	Right.	And that SNR may not be , like , com covering the whole set of S N Rs that you 're getting in the SpeechDat - Car.	false	QMSum_194
Now after cleaning up it 's a different set of SNR.	Right.	And that SNR may not be , like , com covering the whole set of S N Rs that you 're getting in the SpeechDat - Car.	Right , but the SpeechDat - Car data that you 're seeing is also reduced in noise by the noise compensation.	false	QMSum_194
Right.	And that SNR may not be , like , com covering the whole set of S N Rs that you 're getting in the SpeechDat - Car.	Right , but the SpeechDat - Car data that you 're seeing is also reduced in noise by the noise compensation.	Yeah.	false	QMSum_194
And that SNR may not be , like , com covering the whole set of S N Rs that you 're getting in the SpeechDat - Car.	Right , but the SpeechDat - Car data that you 're seeing is also reduced in noise by the noise compensation.	Yeah.	Yeah , yeah , yeah , yeah , it is. But , I 'm saying , there could be some some issues of	false	QMSum_194
Right , but the SpeechDat - Car data that you 're seeing is also reduced in noise by the noise compensation.	Yeah.	Yeah , yeah , yeah , yeah , it is. But , I 'm saying , there could be some some issues of	So.	false	QMSum_194
Yeah.	Yeah , yeah , yeah , yeah , it is. But , I 'm saying , there could be some some issues of	So.	Mm - hmm.	false	QMSum_194
Yeah , yeah , yeah , yeah , it is. But , I 'm saying , there could be some some issues of	So.	Mm - hmm.	Yeah.	false	QMSum_194
So.	Mm - hmm.	Yeah.	Well , if the initial range of SNR is different , we the problem was already there before. And	false	QMSum_194
Mm - hmm.	Yeah.	Well , if the initial range of SNR is different , we the problem was already there before. And	Yeah.	false	QMSum_194
Yeah.	Well , if the initial range of SNR is different , we the problem was already there before. And	Yeah.	Because Mmm	false	QMSum_194
Well , if the initial range of SNR is different , we the problem was already there before. And	Yeah.	Because Mmm	Yeah , I mean , it depends on whether you believe that the noise compensation is equally reducing the noise on the test set and the training set.	false	QMSum_194
Yeah.	Because Mmm	Yeah , I mean , it depends on whether you believe that the noise compensation is equally reducing the noise on the test set and the training set.	Hmm.	false	QMSum_194
Because Mmm	Yeah , I mean , it depends on whether you believe that the noise compensation is equally reducing the noise on the test set and the training set.	Hmm.	Uh	false	QMSum_194
Yeah , I mean , it depends on whether you believe that the noise compensation is equally reducing the noise on the test set and the training set.	Hmm.	Uh	On the test set , yeah.	false	QMSum_194
Hmm.	Uh	On the test set , yeah.	Right ? I mean , you 're saying there 's a mismatch in noise that wasn't there before ,	false	QMSum_194
Uh	On the test set , yeah.	Right ? I mean , you 're saying there 's a mismatch in noise that wasn't there before ,	Hmm. Mm - hmm.	false	QMSum_194
On the test set , yeah.	Right ? I mean , you 're saying there 's a mismatch in noise that wasn't there before ,	Hmm. Mm - hmm.	but if they were both the same before , then if they were both reduic reduced equally , then , there would not be a mismatch.	false	QMSum_194
Right ? I mean , you 're saying there 's a mismatch in noise that wasn't there before ,	Hmm. Mm - hmm.	but if they were both the same before , then if they were both reduic reduced equally , then , there would not be a mismatch.	Mm - hmm.	false	QMSum_194
Hmm. Mm - hmm.	but if they were both the same before , then if they were both reduic reduced equally , then , there would not be a mismatch.	Mm - hmm.	So , I mean , this may be Heaven forbid , this noise compensation process may be imperfect , but. Uh , so maybe it 's treating some things differently.	false	QMSum_194
but if they were both the same before , then if they were both reduic reduced equally , then , there would not be a mismatch.	Mm - hmm.	So , I mean , this may be Heaven forbid , this noise compensation process may be imperfect , but. Uh , so maybe it 's treating some things differently.	Yeah , uh	false	QMSum_194
Mm - hmm.	So , I mean , this may be Heaven forbid , this noise compensation process may be imperfect , but. Uh , so maybe it 's treating some things differently.	Yeah , uh	Well , I I don't know. I I just that could be seen from the TI - digits , uh , testing condition because , um , the noises are from the TI - digits , right ? Noise	false	QMSum_194
So , I mean , this may be Heaven forbid , this noise compensation process may be imperfect , but. Uh , so maybe it 's treating some things differently.	Yeah , uh	Well , I I don't know. I I just that could be seen from the TI - digits , uh , testing condition because , um , the noises are from the TI - digits , right ? Noise	Yeah. So	false	QMSum_194
Yeah , uh	Well , I I don't know. I I just that could be seen from the TI - digits , uh , testing condition because , um , the noises are from the TI - digits , right ? Noise	Yeah. So	So cleaning up the TI - digits and if the performance goes down in the TI - digits mismatch high mismatch like this	false	QMSum_194
Well , I I don't know. I I just that could be seen from the TI - digits , uh , testing condition because , um , the noises are from the TI - digits , right ? Noise	Yeah. So	So cleaning up the TI - digits and if the performance goes down in the TI - digits mismatch high mismatch like this	Clean training , yeah.	false	QMSum_194
Yeah. So	So cleaning up the TI - digits and if the performance goes down in the TI - digits mismatch high mismatch like this	Clean training , yeah.	on a clean training , or zero DB testing.	false	QMSum_194
So cleaning up the TI - digits and if the performance goes down in the TI - digits mismatch high mismatch like this	Clean training , yeah.	on a clean training , or zero DB testing.	Yeah , we 'll so we 'll see. Uh.	false	QMSum_194
Clean training , yeah.	on a clean training , or zero DB testing.	Yeah , we 'll so we 'll see. Uh.	Yeah.	false	QMSum_194
on a clean training , or zero DB testing.	Yeah , we 'll so we 'll see. Uh.	Yeah.	Maybe.	false	QMSum_194
Yeah , we 'll so we 'll see. Uh.	Yeah.	Maybe.	Then it 's something to do.	false	QMSum_194
Yeah.	Maybe.	Then it 's something to do.	Mm - hmm.	false	QMSum_194
Maybe.	Then it 's something to do.	Mm - hmm.	I mean , one of the things about	false	QMSum_194
Then it 's something to do.	Mm - hmm.	I mean , one of the things about	Yeah.	false	QMSum_194
Mm - hmm.	I mean , one of the things about	Yeah.	I mean , the Macrophone data , um , I think , you know , it was recorded over many different telephones.	false	QMSum_194
I mean , one of the things about	Yeah.	I mean , the Macrophone data , um , I think , you know , it was recorded over many different telephones.	Mm - hmm.	false	QMSum_194
Yeah.	I mean , the Macrophone data , um , I think , you know , it was recorded over many different telephones.	Mm - hmm.	And , um , so , there 's lots of different kinds of acoustic conditions. I mean , it 's not artificially added noise or anything. So it 's not the same. I don't think there 's anybody recording over a car from a car , but I think it 's it 's varied enough that if if doing this adjustments , uh , and playing around with it doesn't , uh , make it better , the most uh , it seems like the most obvious thing to do is to improve the training set. Um I mean , what we were uh the condition It it gave us an enormous amount of improvement in what we were doing with Meeting Recorder digits , even though there , again , these m Macrophone digits were very , very different from , uh , what we were going on here. I mean , we weren't talking over a telephone here. But it was just I think just having a a nice variation in acoustic conditions was just a good thing.	false	QMSum_194
I mean , the Macrophone data , um , I think , you know , it was recorded over many different telephones.	Mm - hmm.	And , um , so , there 's lots of different kinds of acoustic conditions. I mean , it 's not artificially added noise or anything. So it 's not the same. I don't think there 's anybody recording over a car from a car , but I think it 's it 's varied enough that if if doing this adjustments , uh , and playing around with it doesn't , uh , make it better , the most uh , it seems like the most obvious thing to do is to improve the training set. Um I mean , what we were uh the condition It it gave us an enormous amount of improvement in what we were doing with Meeting Recorder digits , even though there , again , these m Macrophone digits were very , very different from , uh , what we were going on here. I mean , we weren't talking over a telephone here. But it was just I think just having a a nice variation in acoustic conditions was just a good thing.	Mm - hmm. Yep.	false	QMSum_194
Mm - hmm.	And , um , so , there 's lots of different kinds of acoustic conditions. I mean , it 's not artificially added noise or anything. So it 's not the same. I don't think there 's anybody recording over a car from a car , but I think it 's it 's varied enough that if if doing this adjustments , uh , and playing around with it doesn't , uh , make it better , the most uh , it seems like the most obvious thing to do is to improve the training set. Um I mean , what we were uh the condition It it gave us an enormous amount of improvement in what we were doing with Meeting Recorder digits , even though there , again , these m Macrophone digits were very , very different from , uh , what we were going on here. I mean , we weren't talking over a telephone here. But it was just I think just having a a nice variation in acoustic conditions was just a good thing.	Mm - hmm. Yep.	Mmm.	false	QMSum_194
And , um , so , there 's lots of different kinds of acoustic conditions. I mean , it 's not artificially added noise or anything. So it 's not the same. I don't think there 's anybody recording over a car from a car , but I think it 's it 's varied enough that if if doing this adjustments , uh , and playing around with it doesn't , uh , make it better , the most uh , it seems like the most obvious thing to do is to improve the training set. Um I mean , what we were uh the condition It it gave us an enormous amount of improvement in what we were doing with Meeting Recorder digits , even though there , again , these m Macrophone digits were very , very different from , uh , what we were going on here. I mean , we weren't talking over a telephone here. But it was just I think just having a a nice variation in acoustic conditions was just a good thing.	Mm - hmm. Yep.	Mmm.	Yeah , actually to s eh , what I observed in the HM case is that the number of deletion dramatically increases. It it doubles.	false	QMSum_194
Mm - hmm. Yep.	Mmm.	Yeah , actually to s eh , what I observed in the HM case is that the number of deletion dramatically increases. It it doubles.	Number of deletions.	true	QMSum_194
Mmm.	Yeah , actually to s eh , what I observed in the HM case is that the number of deletion dramatically increases. It it doubles.	Number of deletions.	When I added the num the neural network it doubles the number of deletions. Yeah , so I don't you know how to interpret that , but , mmm	false	QMSum_194
Yeah , actually to s eh , what I observed in the HM case is that the number of deletion dramatically increases. It it doubles.	Number of deletions.	When I added the num the neural network it doubles the number of deletions. Yeah , so I don't you know how to interpret that , but , mmm	Yeah. Me either.	false	QMSum_194
Number of deletions.	When I added the num the neural network it doubles the number of deletions. Yeah , so I don't you know how to interpret that , but , mmm	Yeah. Me either.	And and did an other numbers stay the same ? Insertion substitutions stay the same ?	false	QMSum_194
When I added the num the neural network it doubles the number of deletions. Yeah , so I don't you know how to interpret that , but , mmm	Yeah. Me either.	And and did an other numbers stay the same ? Insertion substitutions stay the same ?	They p stayed the same ,	false	QMSum_194
Yeah. Me either.	And and did an other numbers stay the same ? Insertion substitutions stay the same ?	They p stayed the same ,	Roughly ?	false	QMSum_194
And and did an other numbers stay the same ? Insertion substitutions stay the same ?	They p stayed the same ,	Roughly ?	they maybe they are a little bit uh , lower.	false	QMSum_194
They p stayed the same ,	Roughly ?	they maybe they are a little bit uh , lower.	Uh - huh.	false	QMSum_194
Roughly ?	they maybe they are a little bit uh , lower.	Uh - huh.	They are a little bit better. Yeah. But	false	QMSum_194
they maybe they are a little bit uh , lower.	Uh - huh.	They are a little bit better. Yeah. But	Did they increase the number of deletions even for the cases that got better ?	false	QMSum_194
Uh - huh.	They are a little bit better. Yeah. But	Did they increase the number of deletions even for the cases that got better ?	Mm - hmm.	false	QMSum_194
They are a little bit better. Yeah. But	Did they increase the number of deletions even for the cases that got better ?	Mm - hmm.	Say , for the I mean , it	false	QMSum_194
Did they increase the number of deletions even for the cases that got better ?	Mm - hmm.	Say , for the I mean , it	No , it doesn't.	false	QMSum_194
Mm - hmm.	Say , for the I mean , it	No , it doesn't.	So it 's only the highly mismatched ?	false	QMSum_194
Say , for the I mean , it	No , it doesn't.	So it 's only the highly mismatched ?	No.	false	QMSum_194
No , it doesn't.	So it 's only the highly mismatched ?	No.	And it Remind me again , the " highly mismatched " means that the	false	QMSum_194
So it 's only the highly mismatched ?	No.	And it Remind me again , the " highly mismatched " means that the	Clean training and	false	QMSum_194
No.	And it Remind me again , the " highly mismatched " means that the	Clean training and	Uh , sorry ?	false	QMSum_194
And it Remind me again , the " highly mismatched " means that the	Clean training and	Uh , sorry ?	It 's clean training Well , close microphone training and distant microphone , um , high speed , I think.	false	QMSum_194
Clean training and	Uh , sorry ?	It 's clean training Well , close microphone training and distant microphone , um , high speed , I think.	Close mike training	false	QMSum_194
Uh , sorry ?	It 's clean training Well , close microphone training and distant microphone , um , high speed , I think.	Close mike training	Well The most noisy cases are the distant microphone for testing.	false	QMSum_194
It 's clean training Well , close microphone training and distant microphone , um , high speed , I think.	Close mike training	Well The most noisy cases are the distant microphone for testing.	Right. So Well , maybe the noise subtraction is subtracting off speech.	false	QMSum_194
Close mike training	Well The most noisy cases are the distant microphone for testing.	Right. So Well , maybe the noise subtraction is subtracting off speech.	Separating. Yeah.	false	QMSum_194
Well The most noisy cases are the distant microphone for testing.	Right. So Well , maybe the noise subtraction is subtracting off speech.	Separating. Yeah.	Wh	false	QMSum_194
Right. So Well , maybe the noise subtraction is subtracting off speech.	Separating. Yeah.	Wh	But Yeah. I mean , but without the neural network it 's well , it 's better. It 's just when we add the neural networks.	false	QMSum_194
Separating. Yeah.	Wh	But Yeah. I mean , but without the neural network it 's well , it 's better. It 's just when we add the neural networks.	Yeah , right.	false	QMSum_194
Wh	But Yeah. I mean , but without the neural network it 's well , it 's better. It 's just when we add the neural networks.	Yeah , right.	The feature are the same except that	false	QMSum_194
But Yeah. I mean , but without the neural network it 's well , it 's better. It 's just when we add the neural networks.	Yeah , right.	The feature are the same except that	Uh , that 's right , that 's right. Um	false	QMSum_194
Yeah , right.	The feature are the same except that	Uh , that 's right , that 's right. Um	Well that that says that , you know , the , um the models in in , uh , the recognizer are really paying attention to the neural net features.	false	QMSum_194
The feature are the same except that	Uh , that 's right , that 's right. Um	Well that that says that , you know , the , um the models in in , uh , the recognizer are really paying attention to the neural net features.	Yeah.	false	QMSum_194
Uh , that 's right , that 's right. Um	Well that that says that , you know , the , um the models in in , uh , the recognizer are really paying attention to the neural net features.	Yeah.	Uh.	false	QMSum_194
Well that that says that , you know , the , um the models in in , uh , the recognizer are really paying attention to the neural net features.	Yeah.	Uh.	Mm - hmm.	false	QMSum_194
Yeah.	Uh.	Mm - hmm.	But , yeah , actually the TIMIT noises are sort of a range of noises and they 're not so much the stationary driving kind of noises , right ? It 's it 's pretty different. Isn't it ?	false	QMSum_194

So it 's i All the noises are from the TI - digits ,

Yeah.

right ? So you i

Um They k uh

false

QMSum_194

Yeah.

right ? So you i

Um They k uh

Well , it it 's like the high mismatch of the SpeechDat - Car after cleaning up , maybe having more noise than the the training set of TIMIT after clean s after you do the noise clean - up.

false

QMSum_194

right ? So you i

Um They k uh

Well , it it 's like the high mismatch of the SpeechDat - Car after cleaning up , maybe having more noise than the the training set of TIMIT after clean s after you do the noise clean - up.

Mmm.

false

QMSum_194

Um They k uh

Well , it it 's like the high mismatch of the SpeechDat - Car after cleaning up , maybe having more noise than the the training set of TIMIT after clean s after you do the noise clean - up.

Mmm.

I mean , earlier you never had any compensation , you just trained it straight away.

false

QMSum_194

Well , it it 's like the high mismatch of the SpeechDat - Car after cleaning up , maybe having more noise than the the training set of TIMIT after clean s after you do the noise clean - up.

Mmm.

I mean , earlier you never had any compensation , you just trained it straight away.

Mm - hmm.

false

QMSum_194

Mmm.

I mean , earlier you never had any compensation , you just trained it straight away.

Mm - hmm.

So it had like all these different conditions of S N Rs , actually in their training set of neural net.

false

QMSum_194

I mean , earlier you never had any compensation , you just trained it straight away.

Mm - hmm.

So it had like all these different conditions of S N Rs , actually in their training set of neural net.

Mm - hmm. Mm - hmm.

false

QMSum_194

Mm - hmm.

So it had like all these different conditions of S N Rs , actually in their training set of neural net.

Mm - hmm. Mm - hmm.

But after cleaning up you have now a different set of S N Rs , right ?

false

QMSum_194

So it had like all these different conditions of S N Rs , actually in their training set of neural net.

Mm - hmm. Mm - hmm.

But after cleaning up you have now a different set of S N Rs , right ?

Yeah.

false

QMSum_194

Mm - hmm. Mm - hmm.

But after cleaning up you have now a different set of S N Rs , right ?

Yeah.

For the training of the neural net.

false

QMSum_194

But after cleaning up you have now a different set of S N Rs , right ?

Yeah.

For the training of the neural net.

Mm - hmm.

false

QMSum_194

Yeah.

For the training of the neural net.

Mm - hmm.

And is it something to do with the mismatch that that 's created after the cleaning up , like the high mismatch

false

QMSum_194

For the training of the neural net.

Mm - hmm.

And is it something to do with the mismatch that that 's created after the cleaning up , like the high mismatch

You mean the the most noisy occurrences on SpeechDat - Car might be a lot more noisy than

false

QMSum_194

Mm - hmm.

And is it something to do with the mismatch that that 's created after the cleaning up , like the high mismatch

You mean the the most noisy occurrences on SpeechDat - Car might be a lot more noisy than

Mm - hmm. Of that I mean , the SNR after the noise compensation of the SpeechDat - Car.

false

QMSum_194

And is it something to do with the mismatch that that 's created after the cleaning up , like the high mismatch

You mean the the most noisy occurrences on SpeechDat - Car might be a lot more noisy than

Mm - hmm. Of that I mean , the SNR after the noise compensation of the SpeechDat - Car.

Oh , so Right. So the training the the neural net is being trained with noise compensated stuff.

false

QMSum_194

You mean the the most noisy occurrences on SpeechDat - Car might be a lot more noisy than

Mm - hmm. Of that I mean , the SNR after the noise compensation of the SpeechDat - Car.

Oh , so Right. So the training the the neural net is being trained with noise compensated stuff.

Maybe.

false

QMSum_194

Mm - hmm. Of that I mean , the SNR after the noise compensation of the SpeechDat - Car.

Oh , so Right. So the training the the neural net is being trained with noise compensated stuff.

Maybe.

Yeah.

false

QMSum_194

Oh , so Right. So the training the the neural net is being trained with noise compensated stuff.

Maybe.

Yeah.

Yeah , yeah.

false

QMSum_194

Maybe.

Yeah.

Yeah , yeah.

Which makes sense ,

false

QMSum_194

Yeah.

Yeah , yeah.

Which makes sense ,

Yeah.

false

QMSum_194

Yeah , yeah.

Which makes sense ,

Yeah.

but , uh , you 're saying Yeah , the noisier ones are still going to be , even after our noise compensation , are still gonna be pretty noisy.

false

QMSum_194

Which makes sense ,

Yeah.

but , uh , you 're saying Yeah , the noisier ones are still going to be , even after our noise compensation , are still gonna be pretty noisy.

Yeah.

false

QMSum_194

Yeah.

but , uh , you 're saying Yeah , the noisier ones are still going to be , even after our noise compensation , are still gonna be pretty noisy.

Yeah.

Mm - hmm.

false

QMSum_194

but , uh , you 're saying Yeah , the noisier ones are still going to be , even after our noise compensation , are still gonna be pretty noisy.

Yeah.

Mm - hmm.

Yeah , so now the after - noise compensation the neural net is seeing a different set of S N Rs than that was originally there in the training set. Of TIMIT. Because in the TIMIT it was zero to some clean.

false

QMSum_194

Yeah.

Mm - hmm.

Yeah , so now the after - noise compensation the neural net is seeing a different set of S N Rs than that was originally there in the training set. Of TIMIT. Because in the TIMIT it was zero to some clean.

Right. Yes.

false

QMSum_194

Mm - hmm.

Yeah , so now the after - noise compensation the neural net is seeing a different set of S N Rs than that was originally there in the training set. Of TIMIT. Because in the TIMIT it was zero to some clean.

Right. Yes.

So the net saw all the SNR @ @ conditions.

false

QMSum_194

Yeah , so now the after - noise compensation the neural net is seeing a different set of S N Rs than that was originally there in the training set. Of TIMIT. Because in the TIMIT it was zero to some clean.

Right. Yes.

So the net saw all the SNR @ @ conditions.

Right.

false

QMSum_194

Right. Yes.

So the net saw all the SNR @ @ conditions.

Right.

Now after cleaning up it 's a different set of SNR.

false

QMSum_194

So the net saw all the SNR @ @ conditions.

Right.

Now after cleaning up it 's a different set of SNR.

Right.

false

QMSum_194

Right.

Now after cleaning up it 's a different set of SNR.

Right.

And that SNR may not be , like , com covering the whole set of S N Rs that you 're getting in the SpeechDat - Car.

false

QMSum_194

Now after cleaning up it 's a different set of SNR.

Right.

And that SNR may not be , like , com covering the whole set of S N Rs that you 're getting in the SpeechDat - Car.

Right , but the SpeechDat - Car data that you 're seeing is also reduced in noise by the noise compensation.

false

QMSum_194

Right.

And that SNR may not be , like , com covering the whole set of S N Rs that you 're getting in the SpeechDat - Car.

Right , but the SpeechDat - Car data that you 're seeing is also reduced in noise by the noise compensation.

Yeah.

false

QMSum_194

And that SNR may not be , like , com covering the whole set of S N Rs that you 're getting in the SpeechDat - Car.

Right , but the SpeechDat - Car data that you 're seeing is also reduced in noise by the noise compensation.

Yeah.

Yeah , yeah , yeah , yeah , it is. But , I 'm saying , there could be some some issues of

false

QMSum_194

Right , but the SpeechDat - Car data that you 're seeing is also reduced in noise by the noise compensation.

Yeah.

Yeah , yeah , yeah , yeah , it is. But , I 'm saying , there could be some some issues of

So.

false

QMSum_194

Yeah.

Yeah , yeah , yeah , yeah , it is. But , I 'm saying , there could be some some issues of

So.

Mm - hmm.

false

QMSum_194

Yeah , yeah , yeah , yeah , it is. But , I 'm saying , there could be some some issues of

So.

Mm - hmm.

Yeah.

false

QMSum_194

So.

Mm - hmm.

Yeah.

Well , if the initial range of SNR is different , we the problem was already there before. And

false

QMSum_194

Mm - hmm.

Yeah.

Well , if the initial range of SNR is different , we the problem was already there before. And

Yeah.

false

QMSum_194

Yeah.

Well , if the initial range of SNR is different , we the problem was already there before. And

Yeah.

Because Mmm

false

QMSum_194

Well , if the initial range of SNR is different , we the problem was already there before. And

Yeah.

Because Mmm

Yeah , I mean , it depends on whether you believe that the noise compensation is equally reducing the noise on the test set and the training set.

false

QMSum_194

Yeah.

Because Mmm

Yeah , I mean , it depends on whether you believe that the noise compensation is equally reducing the noise on the test set and the training set.

Hmm.

false

QMSum_194

Because Mmm

Yeah , I mean , it depends on whether you believe that the noise compensation is equally reducing the noise on the test set and the training set.

Hmm.

Uh

false

QMSum_194

Yeah , I mean , it depends on whether you believe that the noise compensation is equally reducing the noise on the test set and the training set.

Hmm.

Uh

On the test set , yeah.

false

QMSum_194

Hmm.

Uh

On the test set , yeah.

Right ? I mean , you 're saying there 's a mismatch in noise that wasn't there before ,

false

QMSum_194

Uh

On the test set , yeah.

Right ? I mean , you 're saying there 's a mismatch in noise that wasn't there before ,

Hmm. Mm - hmm.

false

QMSum_194

On the test set , yeah.

Right ? I mean , you 're saying there 's a mismatch in noise that wasn't there before ,

Hmm. Mm - hmm.

but if they were both the same before , then if they were both reduic reduced equally , then , there would not be a mismatch.

false

QMSum_194

Right ? I mean , you 're saying there 's a mismatch in noise that wasn't there before ,

Hmm. Mm - hmm.

but if they were both the same before , then if they were both reduic reduced equally , then , there would not be a mismatch.

Mm - hmm.

false

QMSum_194

Hmm. Mm - hmm.

but if they were both the same before , then if they were both reduic reduced equally , then , there would not be a mismatch.

Mm - hmm.

So , I mean , this may be Heaven forbid , this noise compensation process may be imperfect , but. Uh , so maybe it 's treating some things differently.

false

QMSum_194

but if they were both the same before , then if they were both reduic reduced equally , then , there would not be a mismatch.

Mm - hmm.

So , I mean , this may be Heaven forbid , this noise compensation process may be imperfect , but. Uh , so maybe it 's treating some things differently.

Yeah , uh

false

QMSum_194

Mm - hmm.

So , I mean , this may be Heaven forbid , this noise compensation process may be imperfect , but. Uh , so maybe it 's treating some things differently.

Yeah , uh

Well , I I don't know. I I just that could be seen from the TI - digits , uh , testing condition because , um , the noises are from the TI - digits , right ? Noise

false

QMSum_194

So , I mean , this may be Heaven forbid , this noise compensation process may be imperfect , but. Uh , so maybe it 's treating some things differently.

Yeah , uh

Well , I I don't know. I I just that could be seen from the TI - digits , uh , testing condition because , um , the noises are from the TI - digits , right ? Noise

Yeah. So

false

QMSum_194

Yeah , uh

Well , I I don't know. I I just that could be seen from the TI - digits , uh , testing condition because , um , the noises are from the TI - digits , right ? Noise

Yeah. So

So cleaning up the TI - digits and if the performance goes down in the TI - digits mismatch high mismatch like this

false

QMSum_194

Well , I I don't know. I I just that could be seen from the TI - digits , uh , testing condition because , um , the noises are from the TI - digits , right ? Noise

Yeah. So

So cleaning up the TI - digits and if the performance goes down in the TI - digits mismatch high mismatch like this

Clean training , yeah.

false

QMSum_194

Yeah. So

So cleaning up the TI - digits and if the performance goes down in the TI - digits mismatch high mismatch like this

Clean training , yeah.

on a clean training , or zero DB testing.

false

QMSum_194

So cleaning up the TI - digits and if the performance goes down in the TI - digits mismatch high mismatch like this

Clean training , yeah.

on a clean training , or zero DB testing.

Yeah , we 'll so we 'll see. Uh.

false

QMSum_194

Clean training , yeah.

on a clean training , or zero DB testing.

Yeah , we 'll so we 'll see. Uh.

Yeah.

false

QMSum_194

on a clean training , or zero DB testing.

Yeah , we 'll so we 'll see. Uh.

Yeah.

Maybe.

false

QMSum_194

Yeah , we 'll so we 'll see. Uh.

Yeah.

Maybe.

Then it 's something to do.

false

QMSum_194

Yeah.

Maybe.

Then it 's something to do.

Mm - hmm.

false

QMSum_194

Maybe.

Then it 's something to do.

Mm - hmm.

I mean , one of the things about

false

QMSum_194

Then it 's something to do.

Mm - hmm.

I mean , one of the things about

Yeah.

false

QMSum_194

Mm - hmm.

I mean , one of the things about

Yeah.

I mean , the Macrophone data , um , I think , you know , it was recorded over many different telephones.

false

QMSum_194

I mean , one of the things about

Yeah.

I mean , the Macrophone data , um , I think , you know , it was recorded over many different telephones.

Mm - hmm.

false

QMSum_194

Yeah.

I mean , the Macrophone data , um , I think , you know , it was recorded over many different telephones.

Mm - hmm.

And , um , so , there 's lots of different kinds of acoustic conditions. I mean , it 's not artificially added noise or anything. So it 's not the same. I don't think there 's anybody recording over a car from a car , but I think it 's it 's varied enough that if if doing this adjustments , uh , and playing around with it doesn't , uh , make it better , the most uh , it seems like the most obvious thing to do is to improve the training set. Um I mean , what we were uh the condition It it gave us an enormous amount of improvement in what we were doing with Meeting Recorder digits , even though there , again , these m Macrophone digits were very , very different from , uh , what we were going on here. I mean , we weren't talking over a telephone here. But it was just I think just having a a nice variation in acoustic conditions was just a good thing.

false

QMSum_194

I mean , the Macrophone data , um , I think , you know , it was recorded over many different telephones.

Mm - hmm.

And , um , so , there 's lots of different kinds of acoustic conditions. I mean , it 's not artificially added noise or anything. So it 's not the same. I don't think there 's anybody recording over a car from a car , but I think it 's it 's varied enough that if if doing this adjustments , uh , and playing around with it doesn't , uh , make it better , the most uh , it seems like the most obvious thing to do is to improve the training set. Um I mean , what we were uh the condition It it gave us an enormous amount of improvement in what we were doing with Meeting Recorder digits , even though there , again , these m Macrophone digits were very , very different from , uh , what we were going on here. I mean , we weren't talking over a telephone here. But it was just I think just having a a nice variation in acoustic conditions was just a good thing.

Mm - hmm. Yep.

false

QMSum_194

Mm - hmm.

And , um , so , there 's lots of different kinds of acoustic conditions. I mean , it 's not artificially added noise or anything. So it 's not the same. I don't think there 's anybody recording over a car from a car , but I think it 's it 's varied enough that if if doing this adjustments , uh , and playing around with it doesn't , uh , make it better , the most uh , it seems like the most obvious thing to do is to improve the training set. Um I mean , what we were uh the condition It it gave us an enormous amount of improvement in what we were doing with Meeting Recorder digits , even though there , again , these m Macrophone digits were very , very different from , uh , what we were going on here. I mean , we weren't talking over a telephone here. But it was just I think just having a a nice variation in acoustic conditions was just a good thing.

Mm - hmm. Yep.

Mmm.

false

QMSum_194

And , um , so , there 's lots of different kinds of acoustic conditions. I mean , it 's not artificially added noise or anything. So it 's not the same. I don't think there 's anybody recording over a car from a car , but I think it 's it 's varied enough that if if doing this adjustments , uh , and playing around with it doesn't , uh , make it better , the most uh , it seems like the most obvious thing to do is to improve the training set. Um I mean , what we were uh the condition It it gave us an enormous amount of improvement in what we were doing with Meeting Recorder digits , even though there , again , these m Macrophone digits were very , very different from , uh , what we were going on here. I mean , we weren't talking over a telephone here. But it was just I think just having a a nice variation in acoustic conditions was just a good thing.

Mm - hmm. Yep.

Mmm.

Yeah , actually to s eh , what I observed in the HM case is that the number of deletion dramatically increases. It it doubles.

false

QMSum_194

Mm - hmm. Yep.

Mmm.

Yeah , actually to s eh , what I observed in the HM case is that the number of deletion dramatically increases. It it doubles.

Number of deletions.

true

QMSum_194

Mmm.

Yeah , actually to s eh , what I observed in the HM case is that the number of deletion dramatically increases. It it doubles.

Number of deletions.

When I added the num the neural network it doubles the number of deletions. Yeah , so I don't you know how to interpret that , but , mmm

false

QMSum_194

Yeah , actually to s eh , what I observed in the HM case is that the number of deletion dramatically increases. It it doubles.

Number of deletions.

When I added the num the neural network it doubles the number of deletions. Yeah , so I don't you know how to interpret that , but , mmm

Yeah. Me either.

false

QMSum_194

Number of deletions.

When I added the num the neural network it doubles the number of deletions. Yeah , so I don't you know how to interpret that , but , mmm

Yeah. Me either.

And and did an other numbers stay the same ? Insertion substitutions stay the same ?

false

QMSum_194

When I added the num the neural network it doubles the number of deletions. Yeah , so I don't you know how to interpret that , but , mmm

Yeah. Me either.

And and did an other numbers stay the same ? Insertion substitutions stay the same ?

They p stayed the same ,

false

QMSum_194

Yeah. Me either.

And and did an other numbers stay the same ? Insertion substitutions stay the same ?

They p stayed the same ,

Roughly ?

false

QMSum_194

And and did an other numbers stay the same ? Insertion substitutions stay the same ?

They p stayed the same ,

Roughly ?

they maybe they are a little bit uh , lower.

false

QMSum_194

They p stayed the same ,

Roughly ?

they maybe they are a little bit uh , lower.

Uh - huh.

false

QMSum_194

Roughly ?

they maybe they are a little bit uh , lower.

Uh - huh.

They are a little bit better. Yeah. But

false

QMSum_194

they maybe they are a little bit uh , lower.

Uh - huh.

They are a little bit better. Yeah. But

Did they increase the number of deletions even for the cases that got better ?

false

QMSum_194

Uh - huh.

They are a little bit better. Yeah. But

Did they increase the number of deletions even for the cases that got better ?

Mm - hmm.

false

QMSum_194

They are a little bit better. Yeah. But

Did they increase the number of deletions even for the cases that got better ?

Mm - hmm.

Say , for the I mean , it

false

QMSum_194

Did they increase the number of deletions even for the cases that got better ?

Mm - hmm.

Say , for the I mean , it

No , it doesn't.

false

QMSum_194

Mm - hmm.

Say , for the I mean , it

No , it doesn't.

So it 's only the highly mismatched ?

false

QMSum_194

Say , for the I mean , it

No , it doesn't.

So it 's only the highly mismatched ?

No.

false

QMSum_194

No , it doesn't.

So it 's only the highly mismatched ?

No.

And it Remind me again , the " highly mismatched " means that the

false

QMSum_194

So it 's only the highly mismatched ?

No.

And it Remind me again , the " highly mismatched " means that the

Clean training and

false

QMSum_194

No.

And it Remind me again , the " highly mismatched " means that the

Clean training and

Uh , sorry ?

false

QMSum_194

And it Remind me again , the " highly mismatched " means that the

Clean training and

Uh , sorry ?

It 's clean training Well , close microphone training and distant microphone , um , high speed , I think.

false

QMSum_194

Clean training and

Uh , sorry ?

It 's clean training Well , close microphone training and distant microphone , um , high speed , I think.

Close mike training

false

QMSum_194

Uh , sorry ?

It 's clean training Well , close microphone training and distant microphone , um , high speed , I think.

Close mike training

Well The most noisy cases are the distant microphone for testing.

false

QMSum_194

It 's clean training Well , close microphone training and distant microphone , um , high speed , I think.

Close mike training

Well The most noisy cases are the distant microphone for testing.

Right. So Well , maybe the noise subtraction is subtracting off speech.

false

QMSum_194

Close mike training

Well The most noisy cases are the distant microphone for testing.

Right. So Well , maybe the noise subtraction is subtracting off speech.

Separating. Yeah.

false

QMSum_194

Well The most noisy cases are the distant microphone for testing.

Right. So Well , maybe the noise subtraction is subtracting off speech.

Separating. Yeah.

Wh

false

QMSum_194

Right. So Well , maybe the noise subtraction is subtracting off speech.

Separating. Yeah.

Wh

But Yeah. I mean , but without the neural network it 's well , it 's better. It 's just when we add the neural networks.

false

QMSum_194

Separating. Yeah.

Wh

But Yeah. I mean , but without the neural network it 's well , it 's better. It 's just when we add the neural networks.

Yeah , right.

false

QMSum_194

Wh

But Yeah. I mean , but without the neural network it 's well , it 's better. It 's just when we add the neural networks.

Yeah , right.

The feature are the same except that

false

QMSum_194

But Yeah. I mean , but without the neural network it 's well , it 's better. It 's just when we add the neural networks.

Yeah , right.

The feature are the same except that

Uh , that 's right , that 's right. Um

false

QMSum_194

Yeah , right.

The feature are the same except that

Uh , that 's right , that 's right. Um

Well that that says that , you know , the , um the models in in , uh , the recognizer are really paying attention to the neural net features.

false

QMSum_194

The feature are the same except that

Uh , that 's right , that 's right. Um

Well that that says that , you know , the , um the models in in , uh , the recognizer are really paying attention to the neural net features.

Yeah.

false

QMSum_194

Uh , that 's right , that 's right. Um

Well that that says that , you know , the , um the models in in , uh , the recognizer are really paying attention to the neural net features.

Yeah.

Uh.

false

QMSum_194

Well that that says that , you know , the , um the models in in , uh , the recognizer are really paying attention to the neural net features.

Yeah.

Uh.

Mm - hmm.

false

QMSum_194

Yeah.

Uh.

Mm - hmm.

But , yeah , actually the TIMIT noises are sort of a range of noises and they 're not so much the stationary driving kind of noises , right ? It 's it 's pretty different. Isn't it ?

false

QMSum_194