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
Right ? In that case you wouldn't necessarily expect it to be better at all.	Oh , yeah , I wasn't necessarily saying it should be better. I 'm just surprised that you 're getting fifteen percent relative worse on the wel	Uh - huh.	But it 's worse.	false	QMSum_194
Oh , yeah , I wasn't necessarily saying it should be better. I 'm just surprised that you 're getting fifteen percent relative worse on the wel	Uh - huh.	But it 's worse.	On the highly mismatched condition.	false	QMSum_194
Uh - huh.	But it 's worse.	On the highly mismatched condition.	On the highly mismatch.	false	QMSum_194
But it 's worse.	On the highly mismatched condition.	On the highly mismatch.	Yeah , I	false	QMSum_194
On the highly mismatched condition.	On the highly mismatch.	Yeah , I	Yeah.	false	QMSum_194
On the highly mismatch.	Yeah , I	Yeah.	So , " highly mismatched condition " means that in fact your training is a bad estimate of your test.	false	QMSum_194
Yeah , I	Yeah.	So , " highly mismatched condition " means that in fact your training is a bad estimate of your test.	Uh - huh.	false	QMSum_194
Yeah.	So , " highly mismatched condition " means that in fact your training is a bad estimate of your test.	Uh - huh.	So having having , uh , a g a l a greater number of features , if they aren't maybe the right features that you use , certainly can e can easily , uh , make things worse. I mean , you 're right. If you have if you have , uh , lots and lots of data , and you have and your your your training is representative of your test , then getting more sources of information should just help. But but it 's It doesn't necessarily work that way.	false	QMSum_194
So , " highly mismatched condition " means that in fact your training is a bad estimate of your test.	Uh - huh.	So having having , uh , a g a l a greater number of features , if they aren't maybe the right features that you use , certainly can e can easily , uh , make things worse. I mean , you 're right. If you have if you have , uh , lots and lots of data , and you have and your your your training is representative of your test , then getting more sources of information should just help. But but it 's It doesn't necessarily work that way.	Huh.	false	QMSum_194
Uh - huh.	So having having , uh , a g a l a greater number of features , if they aren't maybe the right features that you use , certainly can e can easily , uh , make things worse. I mean , you 're right. If you have if you have , uh , lots and lots of data , and you have and your your your training is representative of your test , then getting more sources of information should just help. But but it 's It doesn't necessarily work that way.	Huh.	Mm - hmm.	false	QMSum_194
So having having , uh , a g a l a greater number of features , if they aren't maybe the right features that you use , certainly can e can easily , uh , make things worse. I mean , you 're right. If you have if you have , uh , lots and lots of data , and you have and your your your training is representative of your test , then getting more sources of information should just help. But but it 's It doesn't necessarily work that way.	Huh.	Mm - hmm.	So I wonder , um , Well , what 's your what 's your thought about what to do next with it ?	false	QMSum_194
Huh.	Mm - hmm.	So I wonder , um , Well , what 's your what 's your thought about what to do next with it ?	Um , I don't know. I 'm surprised , because I expected the neural net to help more when there is more mismatch , as it was the case for the	true	QMSum_194
Mm - hmm.	So I wonder , um , Well , what 's your what 's your thought about what to do next with it ?	Um , I don't know. I 'm surprised , because I expected the neural net to help more when there is more mismatch , as it was the case for the	Mm - hmm.	false	QMSum_194
So I wonder , um , Well , what 's your what 's your thought about what to do next with it ?	Um , I don't know. I 'm surprised , because I expected the neural net to help more when there is more mismatch , as it was the case for the	Mm - hmm.	So , was the training set same as the p the February proposal ? OK.	false	QMSum_194
Um , I don't know. I 'm surprised , because I expected the neural net to help more when there is more mismatch , as it was the case for the	Mm - hmm.	So , was the training set same as the p the February proposal ? OK.	Yeah , it 's the same training set , so it 's TIMIT with the TI - digits ' , uh , noises , uh , added.	false	QMSum_194
Mm - hmm.	So , was the training set same as the p the February proposal ? OK.	Yeah , it 's the same training set , so it 's TIMIT with the TI - digits ' , uh , noises , uh , added.	Mm - hmm.	false	QMSum_194
So , was the training set same as the p the February proposal ? OK.	Yeah , it 's the same training set , so it 's TIMIT with the TI - digits ' , uh , noises , uh , added.	Mm - hmm.	Um	false	QMSum_194
Yeah , it 's the same training set , so it 's TIMIT with the TI - digits ' , uh , noises , uh , added.	Mm - hmm.	Um	Well , we might uh , we might have to experiment with , uh better training sets. Again. But ,	false	QMSum_194
Mm - hmm.	Um	Well , we might uh , we might have to experiment with , uh better training sets. Again. But ,	Mm - hmm.	false	QMSum_194
Um	Well , we might uh , we might have to experiment with , uh better training sets. Again. But ,	Mm - hmm.	I The other thing is , I mean , before you found that was the best configuration , but you might have to retest those things now that we have different The rest of it is different , right ? So , um , uh , For instance , what 's the effect of just putting the neural net on without the o other other path ?	false	QMSum_194
Well , we might uh , we might have to experiment with , uh better training sets. Again. But ,	Mm - hmm.	I The other thing is , I mean , before you found that was the best configuration , but you might have to retest those things now that we have different The rest of it is different , right ? So , um , uh , For instance , what 's the effect of just putting the neural net on without the o other other path ?	Mm - hmm.	false	QMSum_194
Mm - hmm.	I The other thing is , I mean , before you found that was the best configuration , but you might have to retest those things now that we have different The rest of it is different , right ? So , um , uh , For instance , what 's the effect of just putting the neural net on without the o other other path ?	Mm - hmm.	I mean , you know what the straight features do.	false	QMSum_194
I The other thing is , I mean , before you found that was the best configuration , but you might have to retest those things now that we have different The rest of it is different , right ? So , um , uh , For instance , what 's the effect of just putting the neural net on without the o other other path ?	Mm - hmm.	I mean , you know what the straight features do.	Yeah.	false	QMSum_194
Mm - hmm.	I mean , you know what the straight features do.	Yeah.	That gives you this. You know what it does in combination.	false	QMSum_194
I mean , you know what the straight features do.	Yeah.	That gives you this. You know what it does in combination.	Mm - hmm.	false	QMSum_194
Yeah.	That gives you this. You know what it does in combination.	Mm - hmm.	You don't necessarily know what	false	QMSum_194
That gives you this. You know what it does in combination.	Mm - hmm.	You don't necessarily know what	What if you did the Would it make sense to do the KLT on the full set of combined features ? Instead of just on the	false	QMSum_194
Mm - hmm.	You don't necessarily know what	What if you did the Would it make sense to do the KLT on the full set of combined features ? Instead of just on the	Yeah. I g I guess. Um. The reason I did it this ways is that in February , it we we tested different things like that , so , having two KLT , having just a KLT for a network , or having a global KLT.	false	QMSum_194
You don't necessarily know what	What if you did the Would it make sense to do the KLT on the full set of combined features ? Instead of just on the	Yeah. I g I guess. Um. The reason I did it this ways is that in February , it we we tested different things like that , so , having two KLT , having just a KLT for a network , or having a global KLT.	Oh , I see.	false	QMSum_194
What if you did the Would it make sense to do the KLT on the full set of combined features ? Instead of just on the	Yeah. I g I guess. Um. The reason I did it this ways is that in February , it we we tested different things like that , so , having two KLT , having just a KLT for a network , or having a global KLT.	Oh , I see.	And	false	QMSum_194
Yeah. I g I guess. Um. The reason I did it this ways is that in February , it we we tested different things like that , so , having two KLT , having just a KLT for a network , or having a global KLT.	Oh , I see.	And	So you tried the global KLT before	false	QMSum_194
Oh , I see.	And	So you tried the global KLT before	Well	false	QMSum_194
And	So you tried the global KLT before	Well	and it didn't really	false	QMSum_194
So you tried the global KLT before	Well	and it didn't really	Yeah. And , uh , th Yeah.	false	QMSum_194
Well	and it didn't really	Yeah. And , uh , th Yeah.	I see.	false	QMSum_194
and it didn't really	Yeah. And , uh , th Yeah.	I see.	The differences between these configurations were not huge , but it was marginally better with this configuration.	false	QMSum_194
Yeah. And , uh , th Yeah.	I see.	The differences between these configurations were not huge , but it was marginally better with this configuration.	Uh - huh. Uh - huh.	false	QMSum_194
I see.	The differences between these configurations were not huge , but it was marginally better with this configuration.	Uh - huh. Uh - huh.	But , yeah , that 's obviously another thing to try ,	false	QMSum_194
The differences between these configurations were not huge , but it was marginally better with this configuration.	Uh - huh. Uh - huh.	But , yeah , that 's obviously another thing to try ,	Um.	false	QMSum_194
Uh - huh. Uh - huh.	But , yeah , that 's obviously another thing to try ,	Um.	since things are things are different.	false	QMSum_194
But , yeah , that 's obviously another thing to try ,	Um.	since things are things are different.	Mm - hmm. Mm - hmm.	false	QMSum_194
Um.	since things are things are different.	Mm - hmm. Mm - hmm.	And I guess if the These are all so all of these seventy - three features are going into , um , the , uh the HMM.	false	QMSum_194
since things are things are different.	Mm - hmm. Mm - hmm.	And I guess if the These are all so all of these seventy - three features are going into , um , the , uh the HMM.	Yeah.	false	QMSum_194
Mm - hmm. Mm - hmm.	And I guess if the These are all so all of these seventy - three features are going into , um , the , uh the HMM.	Yeah.	And is are i i are are any deltas being computed of tha of them ?	false	QMSum_194
And I guess if the These are all so all of these seventy - three features are going into , um , the , uh the HMM.	Yeah.	And is are i i are are any deltas being computed of tha of them ?	Of the straight features , yeah.	false	QMSum_194
Yeah.	And is are i i are are any deltas being computed of tha of them ?	Of the straight features , yeah.	n Not of the	false	QMSum_194
And is are i i are are any deltas being computed of tha of them ?	Of the straight features , yeah.	n Not of the	So. But n th the , um , tandem features are u used as they are.	false	QMSum_194
Of the straight features , yeah.	n Not of the	So. But n th the , um , tandem features are u used as they are.	Are not.	false	QMSum_194
n Not of the	So. But n th the , um , tandem features are u used as they are.	Are not.	So , yeah , maybe we can add some context from these features also as Dan did in in his last work.	false	QMSum_194
So. But n th the , um , tandem features are u used as they are.	Are not.	So , yeah , maybe we can add some context from these features also as Dan did in in his last work.	Could. i Yeah , but the other thing I was thinking was , um Uh , now I lost track of what I was thinking. But.	false	QMSum_194
Are not.	So , yeah , maybe we can add some context from these features also as Dan did in in his last work.	Could. i Yeah , but the other thing I was thinking was , um Uh , now I lost track of what I was thinking. But.	What is the You said there was a limit of sixty features or something ?	false	QMSum_194
So , yeah , maybe we can add some context from these features also as Dan did in in his last work.	Could. i Yeah , but the other thing I was thinking was , um Uh , now I lost track of what I was thinking. But.	What is the You said there was a limit of sixty features or something ?	Mm - hmm.	false	QMSum_194
Could. i Yeah , but the other thing I was thinking was , um Uh , now I lost track of what I was thinking. But.	What is the You said there was a limit of sixty features or something ?	Mm - hmm.	What 's the relation between that limit and the , um , forty - eight uh , forty eight hundred bits per second ?	false	QMSum_194
What is the You said there was a limit of sixty features or something ?	Mm - hmm.	What 's the relation between that limit and the , um , forty - eight uh , forty eight hundred bits per second ?	Oh , I know what I was gonna say.	false	QMSum_194
Mm - hmm.	What 's the relation between that limit and the , um , forty - eight uh , forty eight hundred bits per second ?	Oh , I know what I was gonna say.	Um , not no relation.	false	QMSum_194
What 's the relation between that limit and the , um , forty - eight uh , forty eight hundred bits per second ?	Oh , I know what I was gonna say.	Um , not no relation.	No relation.	false	QMSum_194
Oh , I know what I was gonna say.	Um , not no relation.	No relation.	So I I I don't understand ,	false	QMSum_194
Um , not no relation.	No relation.	So I I I don't understand ,	The f the forty - eight hundred bits is for transmission of some features.	false	QMSum_194
No relation.	So I I I don't understand ,	The f the forty - eight hundred bits is for transmission of some features.	because i I mean , if you 're only using h	false	QMSum_194
So I I I don't understand ,	The f the forty - eight hundred bits is for transmission of some features.	because i I mean , if you 're only using h	And generally , i it s allows you to transmit like , fifteen , uh , cepstrum.	false	QMSum_194
The f the forty - eight hundred bits is for transmission of some features.	because i I mean , if you 're only using h	And generally , i it s allows you to transmit like , fifteen , uh , cepstrum.	The issue was that , um , this is supposed to be a standard that 's then gonna be fed to somebody 's recognizer somewhere which might be , you know , it it might be a concern how many parameters are use u used and so forth. And so , uh , they felt they wanted to set a limit. So they chose sixty. Some people wanted to use hundreds of parameters and and that bothered some other people.	false	QMSum_194
because i I mean , if you 're only using h	And generally , i it s allows you to transmit like , fifteen , uh , cepstrum.	The issue was that , um , this is supposed to be a standard that 's then gonna be fed to somebody 's recognizer somewhere which might be , you know , it it might be a concern how many parameters are use u used and so forth. And so , uh , they felt they wanted to set a limit. So they chose sixty. Some people wanted to use hundreds of parameters and and that bothered some other people.	Uh - huh.	false	QMSum_194
And generally , i it s allows you to transmit like , fifteen , uh , cepstrum.	The issue was that , um , this is supposed to be a standard that 's then gonna be fed to somebody 's recognizer somewhere which might be , you know , it it might be a concern how many parameters are use u used and so forth. And so , uh , they felt they wanted to set a limit. So they chose sixty. Some people wanted to use hundreds of parameters and and that bothered some other people.	Uh - huh.	u And so they just chose that. I I I think it 's kind of r arbitrary too. But but that 's that 's kind of what was chosen. I I remembered what I was going to say. What I was going to say is that , um , maybe maybe with the noise removal , uh , these things are now more correlated. So you have two sets of things that are kind of uncorrelated , uh , within themselves , but they 're pretty correlated with one another.	false	QMSum_194
The issue was that , um , this is supposed to be a standard that 's then gonna be fed to somebody 's recognizer somewhere which might be , you know , it it might be a concern how many parameters are use u used and so forth. And so , uh , they felt they wanted to set a limit. So they chose sixty. Some people wanted to use hundreds of parameters and and that bothered some other people.	Uh - huh.	u And so they just chose that. I I I think it 's kind of r arbitrary too. But but that 's that 's kind of what was chosen. I I remembered what I was going to say. What I was going to say is that , um , maybe maybe with the noise removal , uh , these things are now more correlated. So you have two sets of things that are kind of uncorrelated , uh , within themselves , but they 're pretty correlated with one another.	Mm - hmm.	false	QMSum_194
Uh - huh.	u And so they just chose that. I I I think it 's kind of r arbitrary too. But but that 's that 's kind of what was chosen. I I remembered what I was going to say. What I was going to say is that , um , maybe maybe with the noise removal , uh , these things are now more correlated. So you have two sets of things that are kind of uncorrelated , uh , within themselves , but they 're pretty correlated with one another.	Mm - hmm.	And , um , they 're being fed into these , uh , variants , only Gaussians and so forth , and and , uh ,	false	QMSum_194
u And so they just chose that. I I I think it 's kind of r arbitrary too. But but that 's that 's kind of what was chosen. I I remembered what I was going to say. What I was going to say is that , um , maybe maybe with the noise removal , uh , these things are now more correlated. So you have two sets of things that are kind of uncorrelated , uh , within themselves , but they 're pretty correlated with one another.	Mm - hmm.	And , um , they 're being fed into these , uh , variants , only Gaussians and so forth , and and , uh ,	Mm - hmm.	false	QMSum_194
Mm - hmm.	And , um , they 're being fed into these , uh , variants , only Gaussians and so forth , and and , uh ,	Mm - hmm.	so maybe it would be a better idea now than it was before to , uh , have , uh , one KLT over everything , to de - correlate it.	false	QMSum_194
And , um , they 're being fed into these , uh , variants , only Gaussians and so forth , and and , uh ,	Mm - hmm.	so maybe it would be a better idea now than it was before to , uh , have , uh , one KLT over everything , to de - correlate it.	Mm - hmm. Yeah , I see.	false	QMSum_194
Mm - hmm.	so maybe it would be a better idea now than it was before to , uh , have , uh , one KLT over everything , to de - correlate it.	Mm - hmm. Yeah , I see.	Maybe. You know.	false	QMSum_194
so maybe it would be a better idea now than it was before to , uh , have , uh , one KLT over everything , to de - correlate it.	Mm - hmm. Yeah , I see.	Maybe. You know.	What are the S N Rs in the training set , TIMIT ?	false	QMSum_194
Mm - hmm. Yeah , I see.	Maybe. You know.	What are the S N Rs in the training set , TIMIT ?	It 's , uh , ranging from zero to clean ? Yeah. From zero to clean.	false	QMSum_194
Maybe. You know.	What are the S N Rs in the training set , TIMIT ?	It 's , uh , ranging from zero to clean ? Yeah. From zero to clean.	Mm - hmm.	false	QMSum_194
What are the S N Rs in the training set , TIMIT ?	It 's , uh , ranging from zero to clean ? Yeah. From zero to clean.	Mm - hmm.	Yeah. So we found this this , uh this Macrophone data , and so forth , that we were using for these other experiments , to be pretty good.	false	QMSum_194
It 's , uh , ranging from zero to clean ? Yeah. From zero to clean.	Mm - hmm.	Yeah. So we found this this , uh this Macrophone data , and so forth , that we were using for these other experiments , to be pretty good.	Mm - hmm.	false	QMSum_194
Mm - hmm.	Yeah. So we found this this , uh this Macrophone data , and so forth , that we were using for these other experiments , to be pretty good.	Mm - hmm.	So that 's i after you explore these other alternatives , that might be another way to start looking , is is just improving the training set.	false	QMSum_194
Yeah. So we found this this , uh this Macrophone data , and so forth , that we were using for these other experiments , to be pretty good.	Mm - hmm.	So that 's i after you explore these other alternatives , that might be another way to start looking , is is just improving the training set.	Mm - hmm.	false	QMSum_194
Mm - hmm.	So that 's i after you explore these other alternatives , that might be another way to start looking , is is just improving the training set.	Mm - hmm.	I mean , we were getting , uh , lots better recognition using that , than Of course , you do have the problem that , um , u i we are not able to increase the number of Gaussians , uh , or anything to , uh , uh , to match anything. So we 're only improving the training of our feature set , but that 's still probably something.	false	QMSum_194
So that 's i after you explore these other alternatives , that might be another way to start looking , is is just improving the training set.	Mm - hmm.	I mean , we were getting , uh , lots better recognition using that , than Of course , you do have the problem that , um , u i we are not able to increase the number of Gaussians , uh , or anything to , uh , uh , to match anything. So we 're only improving the training of our feature set , but that 's still probably something.	So you 're saying , add the Macrophone data to the training of the neural net ? The tandem net ?	false	QMSum_194
Mm - hmm.	I mean , we were getting , uh , lots better recognition using that , than Of course , you do have the problem that , um , u i we are not able to increase the number of Gaussians , uh , or anything to , uh , uh , to match anything. So we 're only improving the training of our feature set , but that 's still probably something.	So you 're saying , add the Macrophone data to the training of the neural net ? The tandem net ?	Yeah , that 's the only place that we can train.	false	QMSum_194
I mean , we were getting , uh , lots better recognition using that , than Of course , you do have the problem that , um , u i we are not able to increase the number of Gaussians , uh , or anything to , uh , uh , to match anything. So we 're only improving the training of our feature set , but that 's still probably something.	So you 're saying , add the Macrophone data to the training of the neural net ? The tandem net ?	Yeah , that 's the only place that we can train.	Yeah.	false	QMSum_194
So you 're saying , add the Macrophone data to the training of the neural net ? The tandem net ?	Yeah , that 's the only place that we can train.	Yeah.	We can't train the other stuff with anything other than the standard amount ,	false	QMSum_194
Yeah , that 's the only place that we can train.	Yeah.	We can't train the other stuff with anything other than the standard amount ,	Right.	false	QMSum_194
Yeah.	We can't train the other stuff with anything other than the standard amount ,	Right.	so. Um , um	false	QMSum_194
We can't train the other stuff with anything other than the standard amount ,	Right.	so. Um , um	What what was it trained on again ? The one that you used ?	false	QMSum_194
Right.	so. Um , um	What what was it trained on again ? The one that you used ?	It 's TIMIT with noise.	false	QMSum_194
so. Um , um	What what was it trained on again ? The one that you used ?	It 's TIMIT with noise.	Uh - huh.	false	QMSum_194
What what was it trained on again ? The one that you used ?	It 's TIMIT with noise.	Uh - huh.	Yeah.	false	QMSum_194
It 's TIMIT with noise.	Uh - huh.	Yeah.	So , yeah , it 's rather a small	false	QMSum_194
Uh - huh.	Yeah.	So , yeah , it 's rather a small	How big is the net , by the way ?	false	QMSum_194
Yeah.	So , yeah , it 's rather a small	How big is the net , by the way ?	Um , Uh , it 's , uh , five hundred hidden units. And	false	QMSum_194
So , yeah , it 's rather a small	How big is the net , by the way ?	Um , Uh , it 's , uh , five hundred hidden units. And	And again , you did experiments back then where you made it bigger and it and that was that was sort of the threshold point. Much less than that , it was worse ,	false	QMSum_194
How big is the net , by the way ?	Um , Uh , it 's , uh , five hundred hidden units. And	And again , you did experiments back then where you made it bigger and it and that was that was sort of the threshold point. Much less than that , it was worse ,	Yeah.	false	QMSum_194
Um , Uh , it 's , uh , five hundred hidden units. And	And again , you did experiments back then where you made it bigger and it and that was that was sort of the threshold point. Much less than that , it was worse ,	Yeah.	and	false	QMSum_194
And again , you did experiments back then where you made it bigger and it and that was that was sort of the threshold point. Much less than that , it was worse ,	Yeah.	and	Yeah.	false	QMSum_194
Yeah.	and	Yeah.	much more than that , it wasn't much better. Hmm.	false	QMSum_194
and	Yeah.	much more than that , it wasn't much better. Hmm.	Yeah. @ @ ?	false	QMSum_194
Yeah.	much more than that , it wasn't much better. Hmm.	Yeah. @ @ ?	So is it is it though the performance , big relation in the high ma high mismatch has something to do with the , uh , cleaning up that you that is done on the TIMIT after adding noise ?	false	QMSum_194
much more than that , it wasn't much better. Hmm.	Yeah. @ @ ?	So is it is it though the performance , big relation in the high ma high mismatch has something to do with the , uh , cleaning up that you that is done on the TIMIT after adding noise ?	So it 's i All the noises are from the TI - digits ,	false	QMSum_194
Yeah. @ @ ?	So is it is it though the performance , big relation in the high ma high mismatch has something to do with the , uh , cleaning up that you that is done on the TIMIT after adding noise ?	So it 's i All the noises are from the TI - digits ,	Yeah.	false	QMSum_194
So is it is it though the performance , big relation in the high ma high mismatch has something to do with the , uh , cleaning up that you that is done on the TIMIT after adding noise ?	So it 's i All the noises are from the TI - digits ,	Yeah.	right ? So you i	false	QMSum_194

Right ? In that case you wouldn't necessarily expect it to be better at all.

Oh , yeah , I wasn't necessarily saying it should be better. I 'm just surprised that you 're getting fifteen percent relative worse on the wel

Uh - huh.

But it 's worse.

false

QMSum_194

Oh , yeah , I wasn't necessarily saying it should be better. I 'm just surprised that you 're getting fifteen percent relative worse on the wel

Uh - huh.

But it 's worse.

On the highly mismatched condition.

false

QMSum_194

Uh - huh.

But it 's worse.

On the highly mismatched condition.

On the highly mismatch.

false

QMSum_194

But it 's worse.

On the highly mismatched condition.

On the highly mismatch.

Yeah , I

false

QMSum_194

On the highly mismatched condition.

On the highly mismatch.

Yeah , I

Yeah.

false

QMSum_194

On the highly mismatch.

Yeah , I

Yeah.

So , " highly mismatched condition " means that in fact your training is a bad estimate of your test.

false

QMSum_194

Yeah , I

Yeah.

So , " highly mismatched condition " means that in fact your training is a bad estimate of your test.

Uh - huh.

false

QMSum_194

Yeah.

So , " highly mismatched condition " means that in fact your training is a bad estimate of your test.

Uh - huh.

So having having , uh , a g a l a greater number of features , if they aren't maybe the right features that you use , certainly can e can easily , uh , make things worse. I mean , you 're right. If you have if you have , uh , lots and lots of data , and you have and your your your training is representative of your test , then getting more sources of information should just help. But but it 's It doesn't necessarily work that way.

false

QMSum_194

So , " highly mismatched condition " means that in fact your training is a bad estimate of your test.

Uh - huh.

So having having , uh , a g a l a greater number of features , if they aren't maybe the right features that you use , certainly can e can easily , uh , make things worse. I mean , you 're right. If you have if you have , uh , lots and lots of data , and you have and your your your training is representative of your test , then getting more sources of information should just help. But but it 's It doesn't necessarily work that way.

Huh.

false

QMSum_194

Uh - huh.

So having having , uh , a g a l a greater number of features , if they aren't maybe the right features that you use , certainly can e can easily , uh , make things worse. I mean , you 're right. If you have if you have , uh , lots and lots of data , and you have and your your your training is representative of your test , then getting more sources of information should just help. But but it 's It doesn't necessarily work that way.

Huh.

Mm - hmm.

false

QMSum_194

So having having , uh , a g a l a greater number of features , if they aren't maybe the right features that you use , certainly can e can easily , uh , make things worse. I mean , you 're right. If you have if you have , uh , lots and lots of data , and you have and your your your training is representative of your test , then getting more sources of information should just help. But but it 's It doesn't necessarily work that way.

Huh.

Mm - hmm.

So I wonder , um , Well , what 's your what 's your thought about what to do next with it ?

false

QMSum_194

Huh.

Mm - hmm.

So I wonder , um , Well , what 's your what 's your thought about what to do next with it ?

Um , I don't know. I 'm surprised , because I expected the neural net to help more when there is more mismatch , as it was the case for the

true

QMSum_194

Mm - hmm.

So I wonder , um , Well , what 's your what 's your thought about what to do next with it ?

Um , I don't know. I 'm surprised , because I expected the neural net to help more when there is more mismatch , as it was the case for the

Mm - hmm.

false

QMSum_194

So I wonder , um , Well , what 's your what 's your thought about what to do next with it ?

Um , I don't know. I 'm surprised , because I expected the neural net to help more when there is more mismatch , as it was the case for the

Mm - hmm.

So , was the training set same as the p the February proposal ? OK.

false

QMSum_194

Um , I don't know. I 'm surprised , because I expected the neural net to help more when there is more mismatch , as it was the case for the

Mm - hmm.

So , was the training set same as the p the February proposal ? OK.

Yeah , it 's the same training set , so it 's TIMIT with the TI - digits ' , uh , noises , uh , added.

false

QMSum_194

Mm - hmm.

So , was the training set same as the p the February proposal ? OK.

Yeah , it 's the same training set , so it 's TIMIT with the TI - digits ' , uh , noises , uh , added.

Mm - hmm.

false

QMSum_194

So , was the training set same as the p the February proposal ? OK.

Yeah , it 's the same training set , so it 's TIMIT with the TI - digits ' , uh , noises , uh , added.

Mm - hmm.

Um

false

QMSum_194

Yeah , it 's the same training set , so it 's TIMIT with the TI - digits ' , uh , noises , uh , added.

Mm - hmm.

Um

Well , we might uh , we might have to experiment with , uh better training sets. Again. But ,

false

QMSum_194

Mm - hmm.

Um

Well , we might uh , we might have to experiment with , uh better training sets. Again. But ,

Mm - hmm.

false

QMSum_194

Um

Well , we might uh , we might have to experiment with , uh better training sets. Again. But ,

Mm - hmm.

I The other thing is , I mean , before you found that was the best configuration , but you might have to retest those things now that we have different The rest of it is different , right ? So , um , uh , For instance , what 's the effect of just putting the neural net on without the o other other path ?

false

QMSum_194

Well , we might uh , we might have to experiment with , uh better training sets. Again. But ,

Mm - hmm.

I The other thing is , I mean , before you found that was the best configuration , but you might have to retest those things now that we have different The rest of it is different , right ? So , um , uh , For instance , what 's the effect of just putting the neural net on without the o other other path ?

Mm - hmm.

false

QMSum_194

Mm - hmm.

I The other thing is , I mean , before you found that was the best configuration , but you might have to retest those things now that we have different The rest of it is different , right ? So , um , uh , For instance , what 's the effect of just putting the neural net on without the o other other path ?

Mm - hmm.

I mean , you know what the straight features do.

false

QMSum_194

I The other thing is , I mean , before you found that was the best configuration , but you might have to retest those things now that we have different The rest of it is different , right ? So , um , uh , For instance , what 's the effect of just putting the neural net on without the o other other path ?

Mm - hmm.

I mean , you know what the straight features do.

Yeah.

false

QMSum_194

Mm - hmm.

I mean , you know what the straight features do.

Yeah.

That gives you this. You know what it does in combination.

false

QMSum_194

I mean , you know what the straight features do.

Yeah.

That gives you this. You know what it does in combination.

Mm - hmm.

false

QMSum_194

Yeah.

That gives you this. You know what it does in combination.

Mm - hmm.

You don't necessarily know what

false

QMSum_194

That gives you this. You know what it does in combination.

Mm - hmm.

You don't necessarily know what

What if you did the Would it make sense to do the KLT on the full set of combined features ? Instead of just on the

false

QMSum_194

Mm - hmm.

You don't necessarily know what

What if you did the Would it make sense to do the KLT on the full set of combined features ? Instead of just on the

Yeah. I g I guess. Um. The reason I did it this ways is that in February , it we we tested different things like that , so , having two KLT , having just a KLT for a network , or having a global KLT.

false

QMSum_194

You don't necessarily know what

What if you did the Would it make sense to do the KLT on the full set of combined features ? Instead of just on the

Yeah. I g I guess. Um. The reason I did it this ways is that in February , it we we tested different things like that , so , having two KLT , having just a KLT for a network , or having a global KLT.

Oh , I see.

false

QMSum_194

What if you did the Would it make sense to do the KLT on the full set of combined features ? Instead of just on the

Yeah. I g I guess. Um. The reason I did it this ways is that in February , it we we tested different things like that , so , having two KLT , having just a KLT for a network , or having a global KLT.

Oh , I see.

And

false

QMSum_194

Yeah. I g I guess. Um. The reason I did it this ways is that in February , it we we tested different things like that , so , having two KLT , having just a KLT for a network , or having a global KLT.

Oh , I see.

And

So you tried the global KLT before

false

QMSum_194

Oh , I see.

And

So you tried the global KLT before

Well

false

QMSum_194

And

So you tried the global KLT before

Well

and it didn't really

false

QMSum_194

So you tried the global KLT before

Well

and it didn't really

Yeah. And , uh , th Yeah.

false

QMSum_194

Well

and it didn't really

Yeah. And , uh , th Yeah.

I see.

false

QMSum_194

and it didn't really

Yeah. And , uh , th Yeah.

I see.

The differences between these configurations were not huge , but it was marginally better with this configuration.

false

QMSum_194

Yeah. And , uh , th Yeah.

I see.

The differences between these configurations were not huge , but it was marginally better with this configuration.

Uh - huh. Uh - huh.

false

QMSum_194

I see.

The differences between these configurations were not huge , but it was marginally better with this configuration.

Uh - huh. Uh - huh.

But , yeah , that 's obviously another thing to try ,

false

QMSum_194

The differences between these configurations were not huge , but it was marginally better with this configuration.

Uh - huh. Uh - huh.

But , yeah , that 's obviously another thing to try ,

Um.

false

QMSum_194

Uh - huh. Uh - huh.

But , yeah , that 's obviously another thing to try ,

Um.

since things are things are different.

false

QMSum_194

But , yeah , that 's obviously another thing to try ,

Um.

since things are things are different.

Mm - hmm. Mm - hmm.

false

QMSum_194

Um.

since things are things are different.

Mm - hmm. Mm - hmm.

And I guess if the These are all so all of these seventy - three features are going into , um , the , uh the HMM.

false

QMSum_194

since things are things are different.

Mm - hmm. Mm - hmm.

And I guess if the These are all so all of these seventy - three features are going into , um , the , uh the HMM.

Yeah.

false

QMSum_194

Mm - hmm. Mm - hmm.

And I guess if the These are all so all of these seventy - three features are going into , um , the , uh the HMM.

Yeah.

And is are i i are are any deltas being computed of tha of them ?

false

QMSum_194

And I guess if the These are all so all of these seventy - three features are going into , um , the , uh the HMM.

Yeah.

And is are i i are are any deltas being computed of tha of them ?

Of the straight features , yeah.

false

QMSum_194

Yeah.

And is are i i are are any deltas being computed of tha of them ?

Of the straight features , yeah.

n Not of the

false

QMSum_194

And is are i i are are any deltas being computed of tha of them ?

Of the straight features , yeah.

n Not of the

So. But n th the , um , tandem features are u used as they are.

false

QMSum_194

Of the straight features , yeah.

n Not of the

So. But n th the , um , tandem features are u used as they are.

Are not.

false

QMSum_194

n Not of the

So. But n th the , um , tandem features are u used as they are.

Are not.

So , yeah , maybe we can add some context from these features also as Dan did in in his last work.

false

QMSum_194

So. But n th the , um , tandem features are u used as they are.

Are not.

So , yeah , maybe we can add some context from these features also as Dan did in in his last work.

Could. i Yeah , but the other thing I was thinking was , um Uh , now I lost track of what I was thinking. But.

false

QMSum_194

Are not.

So , yeah , maybe we can add some context from these features also as Dan did in in his last work.

Could. i Yeah , but the other thing I was thinking was , um Uh , now I lost track of what I was thinking. But.

What is the You said there was a limit of sixty features or something ?

false

QMSum_194

So , yeah , maybe we can add some context from these features also as Dan did in in his last work.

Could. i Yeah , but the other thing I was thinking was , um Uh , now I lost track of what I was thinking. But.

What is the You said there was a limit of sixty features or something ?

Mm - hmm.

false

QMSum_194

Could. i Yeah , but the other thing I was thinking was , um Uh , now I lost track of what I was thinking. But.

What is the You said there was a limit of sixty features or something ?

Mm - hmm.

What 's the relation between that limit and the , um , forty - eight uh , forty eight hundred bits per second ?

false

QMSum_194

What is the You said there was a limit of sixty features or something ?

Mm - hmm.

What 's the relation between that limit and the , um , forty - eight uh , forty eight hundred bits per second ?

Oh , I know what I was gonna say.

false

QMSum_194

Mm - hmm.

What 's the relation between that limit and the , um , forty - eight uh , forty eight hundred bits per second ?

Oh , I know what I was gonna say.

Um , not no relation.

false

QMSum_194

What 's the relation between that limit and the , um , forty - eight uh , forty eight hundred bits per second ?

Oh , I know what I was gonna say.

Um , not no relation.

No relation.

false

QMSum_194

Oh , I know what I was gonna say.

Um , not no relation.

No relation.

So I I I don't understand ,

false

QMSum_194

Um , not no relation.

No relation.

So I I I don't understand ,

The f the forty - eight hundred bits is for transmission of some features.

false

QMSum_194

No relation.

So I I I don't understand ,

The f the forty - eight hundred bits is for transmission of some features.

because i I mean , if you 're only using h

false

QMSum_194

So I I I don't understand ,

The f the forty - eight hundred bits is for transmission of some features.

because i I mean , if you 're only using h

And generally , i it s allows you to transmit like , fifteen , uh , cepstrum.

false

QMSum_194

The f the forty - eight hundred bits is for transmission of some features.

because i I mean , if you 're only using h

And generally , i it s allows you to transmit like , fifteen , uh , cepstrum.

The issue was that , um , this is supposed to be a standard that 's then gonna be fed to somebody 's recognizer somewhere which might be , you know , it it might be a concern how many parameters are use u used and so forth. And so , uh , they felt they wanted to set a limit. So they chose sixty. Some people wanted to use hundreds of parameters and and that bothered some other people.

false

QMSum_194

because i I mean , if you 're only using h

And generally , i it s allows you to transmit like , fifteen , uh , cepstrum.

The issue was that , um , this is supposed to be a standard that 's then gonna be fed to somebody 's recognizer somewhere which might be , you know , it it might be a concern how many parameters are use u used and so forth. And so , uh , they felt they wanted to set a limit. So they chose sixty. Some people wanted to use hundreds of parameters and and that bothered some other people.

Uh - huh.

false

QMSum_194

And generally , i it s allows you to transmit like , fifteen , uh , cepstrum.

The issue was that , um , this is supposed to be a standard that 's then gonna be fed to somebody 's recognizer somewhere which might be , you know , it it might be a concern how many parameters are use u used and so forth. And so , uh , they felt they wanted to set a limit. So they chose sixty. Some people wanted to use hundreds of parameters and and that bothered some other people.

Uh - huh.

u And so they just chose that. I I I think it 's kind of r arbitrary too. But but that 's that 's kind of what was chosen. I I remembered what I was going to say. What I was going to say is that , um , maybe maybe with the noise removal , uh , these things are now more correlated. So you have two sets of things that are kind of uncorrelated , uh , within themselves , but they 're pretty correlated with one another.

false

QMSum_194

The issue was that , um , this is supposed to be a standard that 's then gonna be fed to somebody 's recognizer somewhere which might be , you know , it it might be a concern how many parameters are use u used and so forth. And so , uh , they felt they wanted to set a limit. So they chose sixty. Some people wanted to use hundreds of parameters and and that bothered some other people.

Uh - huh.

u And so they just chose that. I I I think it 's kind of r arbitrary too. But but that 's that 's kind of what was chosen. I I remembered what I was going to say. What I was going to say is that , um , maybe maybe with the noise removal , uh , these things are now more correlated. So you have two sets of things that are kind of uncorrelated , uh , within themselves , but they 're pretty correlated with one another.

Mm - hmm.

false

QMSum_194

Uh - huh.

u And so they just chose that. I I I think it 's kind of r arbitrary too. But but that 's that 's kind of what was chosen. I I remembered what I was going to say. What I was going to say is that , um , maybe maybe with the noise removal , uh , these things are now more correlated. So you have two sets of things that are kind of uncorrelated , uh , within themselves , but they 're pretty correlated with one another.

Mm - hmm.

And , um , they 're being fed into these , uh , variants , only Gaussians and so forth , and and , uh ,

false

QMSum_194

u And so they just chose that. I I I think it 's kind of r arbitrary too. But but that 's that 's kind of what was chosen. I I remembered what I was going to say. What I was going to say is that , um , maybe maybe with the noise removal , uh , these things are now more correlated. So you have two sets of things that are kind of uncorrelated , uh , within themselves , but they 're pretty correlated with one another.

Mm - hmm.

And , um , they 're being fed into these , uh , variants , only Gaussians and so forth , and and , uh ,

Mm - hmm.

false

QMSum_194

Mm - hmm.

And , um , they 're being fed into these , uh , variants , only Gaussians and so forth , and and , uh ,

Mm - hmm.

so maybe it would be a better idea now than it was before to , uh , have , uh , one KLT over everything , to de - correlate it.

false

QMSum_194

And , um , they 're being fed into these , uh , variants , only Gaussians and so forth , and and , uh ,

Mm - hmm.

so maybe it would be a better idea now than it was before to , uh , have , uh , one KLT over everything , to de - correlate it.

Mm - hmm. Yeah , I see.

false

QMSum_194

Mm - hmm.

so maybe it would be a better idea now than it was before to , uh , have , uh , one KLT over everything , to de - correlate it.

Mm - hmm. Yeah , I see.

Maybe. You know.

false

QMSum_194

so maybe it would be a better idea now than it was before to , uh , have , uh , one KLT over everything , to de - correlate it.

Mm - hmm. Yeah , I see.

Maybe. You know.

What are the S N Rs in the training set , TIMIT ?

false

QMSum_194

Mm - hmm. Yeah , I see.

Maybe. You know.

What are the S N Rs in the training set , TIMIT ?

It 's , uh , ranging from zero to clean ? Yeah. From zero to clean.

false

QMSum_194

Maybe. You know.

What are the S N Rs in the training set , TIMIT ?

It 's , uh , ranging from zero to clean ? Yeah. From zero to clean.

Mm - hmm.

false

QMSum_194

What are the S N Rs in the training set , TIMIT ?

It 's , uh , ranging from zero to clean ? Yeah. From zero to clean.

Mm - hmm.

Yeah. So we found this this , uh this Macrophone data , and so forth , that we were using for these other experiments , to be pretty good.

false

QMSum_194

It 's , uh , ranging from zero to clean ? Yeah. From zero to clean.

Mm - hmm.

Yeah. So we found this this , uh this Macrophone data , and so forth , that we were using for these other experiments , to be pretty good.

Mm - hmm.

false

QMSum_194

Mm - hmm.

Yeah. So we found this this , uh this Macrophone data , and so forth , that we were using for these other experiments , to be pretty good.

Mm - hmm.

So that 's i after you explore these other alternatives , that might be another way to start looking , is is just improving the training set.

false

QMSum_194

Yeah. So we found this this , uh this Macrophone data , and so forth , that we were using for these other experiments , to be pretty good.

Mm - hmm.

So that 's i after you explore these other alternatives , that might be another way to start looking , is is just improving the training set.

Mm - hmm.

false

QMSum_194

Mm - hmm.

So that 's i after you explore these other alternatives , that might be another way to start looking , is is just improving the training set.

Mm - hmm.

I mean , we were getting , uh , lots better recognition using that , than Of course , you do have the problem that , um , u i we are not able to increase the number of Gaussians , uh , or anything to , uh , uh , to match anything. So we 're only improving the training of our feature set , but that 's still probably something.

false

QMSum_194

So that 's i after you explore these other alternatives , that might be another way to start looking , is is just improving the training set.

Mm - hmm.

I mean , we were getting , uh , lots better recognition using that , than Of course , you do have the problem that , um , u i we are not able to increase the number of Gaussians , uh , or anything to , uh , uh , to match anything. So we 're only improving the training of our feature set , but that 's still probably something.

So you 're saying , add the Macrophone data to the training of the neural net ? The tandem net ?

false

QMSum_194

Mm - hmm.

I mean , we were getting , uh , lots better recognition using that , than Of course , you do have the problem that , um , u i we are not able to increase the number of Gaussians , uh , or anything to , uh , uh , to match anything. So we 're only improving the training of our feature set , but that 's still probably something.

So you 're saying , add the Macrophone data to the training of the neural net ? The tandem net ?

Yeah , that 's the only place that we can train.

false

QMSum_194

I mean , we were getting , uh , lots better recognition using that , than Of course , you do have the problem that , um , u i we are not able to increase the number of Gaussians , uh , or anything to , uh , uh , to match anything. So we 're only improving the training of our feature set , but that 's still probably something.

So you 're saying , add the Macrophone data to the training of the neural net ? The tandem net ?

Yeah , that 's the only place that we can train.

Yeah.

false

QMSum_194

So you 're saying , add the Macrophone data to the training of the neural net ? The tandem net ?

Yeah , that 's the only place that we can train.

Yeah.

We can't train the other stuff with anything other than the standard amount ,

false

QMSum_194

Yeah , that 's the only place that we can train.

Yeah.

We can't train the other stuff with anything other than the standard amount ,

Right.

false

QMSum_194

Yeah.

We can't train the other stuff with anything other than the standard amount ,

Right.

so. Um , um

false

QMSum_194

We can't train the other stuff with anything other than the standard amount ,

Right.

so. Um , um

What what was it trained on again ? The one that you used ?

false

QMSum_194

Right.

so. Um , um

What what was it trained on again ? The one that you used ?

It 's TIMIT with noise.

false

QMSum_194

so. Um , um

What what was it trained on again ? The one that you used ?

It 's TIMIT with noise.

Uh - huh.

false

QMSum_194

What what was it trained on again ? The one that you used ?

It 's TIMIT with noise.

Uh - huh.

Yeah.

false

QMSum_194

It 's TIMIT with noise.

Uh - huh.

Yeah.

So , yeah , it 's rather a small

false

QMSum_194

Uh - huh.

Yeah.

So , yeah , it 's rather a small

How big is the net , by the way ?

false

QMSum_194

Yeah.

So , yeah , it 's rather a small

How big is the net , by the way ?

Um , Uh , it 's , uh , five hundred hidden units. And

false

QMSum_194

So , yeah , it 's rather a small

How big is the net , by the way ?

Um , Uh , it 's , uh , five hundred hidden units. And

And again , you did experiments back then where you made it bigger and it and that was that was sort of the threshold point. Much less than that , it was worse ,

false

QMSum_194

How big is the net , by the way ?

Um , Uh , it 's , uh , five hundred hidden units. And

And again , you did experiments back then where you made it bigger and it and that was that was sort of the threshold point. Much less than that , it was worse ,

Yeah.

false

QMSum_194

Um , Uh , it 's , uh , five hundred hidden units. And

And again , you did experiments back then where you made it bigger and it and that was that was sort of the threshold point. Much less than that , it was worse ,

Yeah.

and

false

QMSum_194

And again , you did experiments back then where you made it bigger and it and that was that was sort of the threshold point. Much less than that , it was worse ,

Yeah.

and

Yeah.

false

QMSum_194

Yeah.

and

Yeah.

much more than that , it wasn't much better. Hmm.

false

QMSum_194

and

Yeah.

much more than that , it wasn't much better. Hmm.

Yeah. @ @ ?

false

QMSum_194

Yeah.

much more than that , it wasn't much better. Hmm.

Yeah. @ @ ?

So is it is it though the performance , big relation in the high ma high mismatch has something to do with the , uh , cleaning up that you that is done on the TIMIT after adding noise ?

false

QMSum_194

much more than that , it wasn't much better. Hmm.

Yeah. @ @ ?

So is it is it though the performance , big relation in the high ma high mismatch has something to do with the , uh , cleaning up that you that is done on the TIMIT after adding noise ?

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

false

QMSum_194

Yeah. @ @ ?

So is it is it though the performance , big relation in the high ma high mismatch has something to do with the , uh , cleaning up that you that is done on the TIMIT after adding noise ?

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

Yeah.

false

QMSum_194

So is it is it though the performance , big relation in the high ma high mismatch has something to do with the , uh , cleaning up that you that is done on the TIMIT after adding noise ?

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

Yeah.

right ? So you i

false

QMSum_194