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
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 ?	Uh , there is a car noise. So there are f just four noises. Um , uh , " Car " , I think , " Babble " ,	false	QMSum_194
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 ?	Uh , there is a car noise. So there are f just four noises. Um , uh , " Car " , I think , " Babble " ,	" Babble. "	false	QMSum_194
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 ?	Uh , there is a car noise. So there are f just four noises. Um , uh , " Car " , I think , " Babble " ,	" Babble. "	" Subway " , right ? and	false	QMSum_194
Uh , there is a car noise. So there are f just four noises. Um , uh , " Car " , I think , " Babble " ,	" Babble. "	" Subway " , right ? and	" Street " or " Airport " or something.	false	QMSum_194
" Babble. "	" Subway " , right ? and	" Street " or " Airport " or something.	and " Street " isn't	false	QMSum_194
" Subway " , right ? and	" Street " or " Airport " or something.	and " Street " isn't	Or " Train station ".	false	QMSum_194
" Street " or " Airport " or something.	and " Street " isn't	Or " Train station ".	" Train station " , yeah.	false	QMSum_194
and " Street " isn't	Or " Train station ".	" Train station " , yeah.	Yeah.	false	QMSum_194
Or " Train station ".	" Train station " , yeah.	Yeah.	So it 's mostly Well , " Car " is stationary ,	false	QMSum_194
" Train station " , yeah.	Yeah.	So it 's mostly Well , " Car " is stationary ,	Mm - hmm.	false	QMSum_194
Yeah.	So it 's mostly Well , " Car " is stationary ,	Mm - hmm.	" Babble " , it 's a stationary background plus some voices ,	false	QMSum_194
So it 's mostly Well , " Car " is stationary ,	Mm - hmm.	" Babble " , it 's a stationary background plus some voices ,	Mm - hmm.	false	QMSum_194
Mm - hmm.	" Babble " , it 's a stationary background plus some voices ,	Mm - hmm.	some speech over it. And the other two are rather stationary also.	false	QMSum_194
" Babble " , it 's a stationary background plus some voices ,	Mm - hmm.	some speech over it. And the other two are rather stationary also.	Well , I I think that if you run it Actually , you maybe you remember this. When you in in the old experiments when you ran with the neural net only , and didn't have this side path , um , uh , with the the pure features as well , did it make things better to have the neural net ?	false	QMSum_194
Mm - hmm.	some speech over it. And the other two are rather stationary also.	Well , I I think that if you run it Actually , you maybe you remember this. When you in in the old experiments when you ran with the neural net only , and didn't have this side path , um , uh , with the the pure features as well , did it make things better to have the neural net ?	Mm - hmm.	false	QMSum_194
some speech over it. And the other two are rather stationary also.	Well , I I think that if you run it Actually , you maybe you remember this. When you in in the old experiments when you ran with the neural net only , and didn't have this side path , um , uh , with the the pure features as well , did it make things better to have the neural net ?	Mm - hmm.	Was it about the same ? Uh , w i	false	QMSum_194
Well , I I think that if you run it Actually , you maybe you remember this. When you in in the old experiments when you ran with the neural net only , and didn't have this side path , um , uh , with the the pure features as well , did it make things better to have the neural net ?	Mm - hmm.	Was it about the same ? Uh , w i	It was b a little bit worse.	false	QMSum_194
Mm - hmm.	Was it about the same ? Uh , w i	It was b a little bit worse.	Than ?	false	QMSum_194
Was it about the same ? Uh , w i	It was b a little bit worse.	Than ?	Than just the features , yeah.	false	QMSum_194
It was b a little bit worse.	Than ?	Than just the features , yeah.	So , until you put the second path in with the pure features , the neural net wasn't helping at all.	false	QMSum_194
Than ?	Than just the features , yeah.	So , until you put the second path in with the pure features , the neural net wasn't helping at all.	Mm - hmm.	false	QMSum_194
Than just the features , yeah.	So , until you put the second path in with the pure features , the neural net wasn't helping at all.	Mm - hmm.	Well , that 's interesting.	false	QMSum_194
So , until you put the second path in with the pure features , the neural net wasn't helping at all.	Mm - hmm.	Well , that 's interesting.	It was helping , uh , if the features are b were bad ,	false	QMSum_194
Mm - hmm.	Well , that 's interesting.	It was helping , uh , if the features are b were bad ,	Yeah.	false	QMSum_194
Well , that 's interesting.	It was helping , uh , if the features are b were bad ,	Yeah.	I mean. Just plain P L Ps or M F	false	QMSum_194
It was helping , uh , if the features are b were bad ,	Yeah.	I mean. Just plain P L Ps or M F	Yeah.	false	QMSum_194
Yeah.	I mean. Just plain P L Ps or M F	Yeah.	C Cs. as soon as we added LDA on - line normalization , and all these things , then	false	QMSum_194
I mean. Just plain P L Ps or M F	Yeah.	C Cs. as soon as we added LDA on - line normalization , and all these things , then	They were doing similar enough things. Well , I still think it would be k sort of interesting to see what would happen if you just had the neural net without the side thing.	false	QMSum_194
Yeah.	C Cs. as soon as we added LDA on - line normalization , and all these things , then	They were doing similar enough things. Well , I still think it would be k sort of interesting to see what would happen if you just had the neural net without the side thing.	Yeah ,	false	QMSum_194
C Cs. as soon as we added LDA on - line normalization , and all these things , then	They were doing similar enough things. Well , I still think it would be k sort of interesting to see what would happen if you just had the neural net without the side thing.	Yeah ,	And and the thing I I have in mind is , uh , maybe you 'll see that the results are not just a little bit worse.	false	QMSum_194
They were doing similar enough things. Well , I still think it would be k sort of interesting to see what would happen if you just had the neural net without the side thing.	Yeah ,	And and the thing I I have in mind is , uh , maybe you 'll see that the results are not just a little bit worse.	mm - hmm.	false	QMSum_194
Yeah ,	And and the thing I I have in mind is , uh , maybe you 'll see that the results are not just a little bit worse.	mm - hmm.	Maybe that they 're a lot worse. You know ? And , um But if on the ha other hand , uh , it 's , say , somewhere in between what you 're seeing now and and and , uh , what you 'd have with just the pure features , then maybe there is some problem of a of a , uh , combination of these things , or correlation between them somehow.	false	QMSum_194
And and the thing I I have in mind is , uh , maybe you 'll see that the results are not just a little bit worse.	mm - hmm.	Maybe that they 're a lot worse. You know ? And , um But if on the ha other hand , uh , it 's , say , somewhere in between what you 're seeing now and and and , uh , what you 'd have with just the pure features , then maybe there is some problem of a of a , uh , combination of these things , or correlation between them somehow.	Mm - hmm.	false	QMSum_194
mm - hmm.	Maybe that they 're a lot worse. You know ? And , um But if on the ha other hand , uh , it 's , say , somewhere in between what you 're seeing now and and and , uh , what you 'd have with just the pure features , then maybe there is some problem of a of a , uh , combination of these things , or correlation between them somehow.	Mm - hmm.	If it really is that the net is hurting you at the moment , then I think the issue is to focus on on , uh , improving the the net.	false	QMSum_194
Maybe that they 're a lot worse. You know ? And , um But if on the ha other hand , uh , it 's , say , somewhere in between what you 're seeing now and and and , uh , what you 'd have with just the pure features , then maybe there is some problem of a of a , uh , combination of these things , or correlation between them somehow.	Mm - hmm.	If it really is that the net is hurting you at the moment , then I think the issue is to focus on on , uh , improving the the net.	Yeah ,	false	QMSum_194
Mm - hmm.	If it really is that the net is hurting you at the moment , then I think the issue is to focus on on , uh , improving the the net.	Yeah ,	Um.	false	QMSum_194
If it really is that the net is hurting you at the moment , then I think the issue is to focus on on , uh , improving the the net.	Yeah ,	Um.	mm - hmm.	false	QMSum_194
Yeah ,	Um.	mm - hmm.	So what 's the overall effe I mean , you haven't done all the experiments but you said it was i somewhat better , say , five percent better , for the first two conditions , and fifteen percent worse for the other one ? But it 's but of course that one 's weighted lower ,	false	QMSum_194
Um.	mm - hmm.	So what 's the overall effe I mean , you haven't done all the experiments but you said it was i somewhat better , say , five percent better , for the first two conditions , and fifteen percent worse for the other one ? But it 's but of course that one 's weighted lower ,	Y yeah , oh. Yeah.	false	QMSum_194
mm - hmm.	So what 's the overall effe I mean , you haven't done all the experiments but you said it was i somewhat better , say , five percent better , for the first two conditions , and fifteen percent worse for the other one ? But it 's but of course that one 's weighted lower ,	Y yeah , oh. Yeah.	so I wonder what the net effect is.	false	QMSum_194
So what 's the overall effe I mean , you haven't done all the experiments but you said it was i somewhat better , say , five percent better , for the first two conditions , and fifteen percent worse for the other one ? But it 's but of course that one 's weighted lower ,	Y yeah , oh. Yeah.	so I wonder what the net effect is.	I d I I think it 's it was one or two percent. That 's not that bad , but it was l like two percent relative worse on SpeechDat - Car. I have to to check that. Well , I have I will.	false	QMSum_194
Y yeah , oh. Yeah.	so I wonder what the net effect is.	I d I I think it 's it was one or two percent. That 's not that bad , but it was l like two percent relative worse on SpeechDat - Car. I have to to check that. Well , I have I will.	Well , it will overall it will be still better even if it is fifteen percent worse , because the fifteen percent worse is given like f w twenty - five point two five eight.	false	QMSum_194
so I wonder what the net effect is.	I d I I think it 's it was one or two percent. That 's not that bad , but it was l like two percent relative worse on SpeechDat - Car. I have to to check that. Well , I have I will.	Well , it will overall it will be still better even if it is fifteen percent worse , because the fifteen percent worse is given like f w twenty - five point two five eight.	Right.	false	QMSum_194
I d I I think it 's it was one or two percent. That 's not that bad , but it was l like two percent relative worse on SpeechDat - Car. I have to to check that. Well , I have I will.	Well , it will overall it will be still better even if it is fifteen percent worse , because the fifteen percent worse is given like f w twenty - five point two five eight.	Right.	Mm - hmm. Hmm.	false	QMSum_194
Well , it will overall it will be still better even if it is fifteen percent worse , because the fifteen percent worse is given like f w twenty - five point two five eight.	Right.	Mm - hmm. Hmm.	Right. So the so the worst it could be , if the others were exactly the same , is four ,	false	QMSum_194
Right.	Mm - hmm. Hmm.	Right. So the so the worst it could be , if the others were exactly the same , is four ,	Is it like	false	QMSum_194
Mm - hmm. Hmm.	Right. So the so the worst it could be , if the others were exactly the same , is four ,	Is it like	and and , uh , in fact since the others are somewhat better	false	QMSum_194
Right. So the so the worst it could be , if the others were exactly the same , is four ,	Is it like	and and , uh , in fact since the others are somewhat better	Yeah , so it 's four. Is i So either it 'll get cancelled out , or you 'll get , like , almost the same.	false	QMSum_194
Is it like	and and , uh , in fact since the others are somewhat better	Yeah , so it 's four. Is i So either it 'll get cancelled out , or you 'll get , like , almost the same.	Uh.	false	QMSum_194
and and , uh , in fact since the others are somewhat better	Yeah , so it 's four. Is i So either it 'll get cancelled out , or you 'll get , like , almost the same.	Uh.	Yeah , it was it was slightly worse.	false	QMSum_194
Yeah , so it 's four. Is i So either it 'll get cancelled out , or you 'll get , like , almost the same.	Uh.	Yeah , it was it was slightly worse.	Slightly bad. Yeah.	false	QMSum_194
Uh.	Yeah , it was it was slightly worse.	Slightly bad. Yeah.	Um ,	false	QMSum_194
Yeah , it was it was slightly worse.	Slightly bad. Yeah.	Um ,	Yeah , it should be pretty close to cancelled out.	false	QMSum_194
Slightly bad. Yeah.	Um ,	Yeah , it should be pretty close to cancelled out.	Yeah.	false	QMSum_194
Um ,	Yeah , it should be pretty close to cancelled out.	Yeah.	You know , I 've been wondering about something.	false	QMSum_194
Yeah , it should be pretty close to cancelled out.	Yeah.	You know , I 've been wondering about something.	Mm - hmm.	false	QMSum_194
Yeah.	You know , I 've been wondering about something.	Mm - hmm.	In the , um a lot of the , um the Hub - five systems , um , recently have been using LDA. and and they , um They run LDA on the features right before they train the models. So there 's the the LDA is is right there before the H M	false	QMSum_194
You know , I 've been wondering about something.	Mm - hmm.	In the , um a lot of the , um the Hub - five systems , um , recently have been using LDA. and and they , um They run LDA on the features right before they train the models. So there 's the the LDA is is right there before the H M	Yeah.	false	QMSum_194
Mm - hmm.	In the , um a lot of the , um the Hub - five systems , um , recently have been using LDA. and and they , um They run LDA on the features right before they train the models. So there 's the the LDA is is right there before the H M	Yeah.	So , you guys are using LDA but it seems like it 's pretty far back in the process.	false	QMSum_194
In the , um a lot of the , um the Hub - five systems , um , recently have been using LDA. and and they , um They run LDA on the features right before they train the models. So there 's the the LDA is is right there before the H M	Yeah.	So , you guys are using LDA but it seems like it 's pretty far back in the process.	Uh , this LDA is different from the LDA that you are talking about. The LDA that you saying is , like , you take a block of features , like nine frames or something , and then do an LDA on it ,	false	QMSum_194
Yeah.	So , you guys are using LDA but it seems like it 's pretty far back in the process.	Uh , this LDA is different from the LDA that you are talking about. The LDA that you saying is , like , you take a block of features , like nine frames or something , and then do an LDA on it ,	Yeah. Uh - huh.	false	QMSum_194
So , you guys are using LDA but it seems like it 's pretty far back in the process.	Uh , this LDA is different from the LDA that you are talking about. The LDA that you saying is , like , you take a block of features , like nine frames or something , and then do an LDA on it ,	Yeah. Uh - huh.	and then reduce the dimensionality to something like twenty - four or something like that.	false	QMSum_194
Uh , this LDA is different from the LDA that you are talking about. The LDA that you saying is , like , you take a block of features , like nine frames or something , and then do an LDA on it ,	Yeah. Uh - huh.	and then reduce the dimensionality to something like twenty - four or something like that.	Yeah , you c you c you can.	false	QMSum_194
Yeah. Uh - huh.	and then reduce the dimensionality to something like twenty - four or something like that.	Yeah , you c you c you can.	And then feed it to HMM.	false	QMSum_194
and then reduce the dimensionality to something like twenty - four or something like that.	Yeah , you c you c you can.	And then feed it to HMM.	I mean , it 's you know , you 're just basically i	false	QMSum_194
Yeah , you c you c you can.	And then feed it to HMM.	I mean , it 's you know , you 're just basically i	Yeah , so this is like a two d two dimensional tile.	false	QMSum_194
And then feed it to HMM.	I mean , it 's you know , you 're just basically i	Yeah , so this is like a two d two dimensional tile.	You 're shifting the feature space. Yeah.	false	QMSum_194
I mean , it 's you know , you 're just basically i	Yeah , so this is like a two d two dimensional tile.	You 're shifting the feature space. Yeah.	So this is a two dimensional tile. And the LDA that we are f applying is only in time , not in frequency high cost frequency. So it 's like more like a filtering in time , rather than doing a r	false	QMSum_194
Yeah , so this is like a two d two dimensional tile.	You 're shifting the feature space. Yeah.	So this is a two dimensional tile. And the LDA that we are f applying is only in time , not in frequency high cost frequency. So it 's like more like a filtering in time , rather than doing a r	Ah. OK. So what i what about , um i u what i w I mean , I don't know if this is a good idea or not , but what if you put ran the other kind of LDA , uh , on your features right before they go into the HMM ?	false	QMSum_194
You 're shifting the feature space. Yeah.	So this is a two dimensional tile. And the LDA that we are f applying is only in time , not in frequency high cost frequency. So it 's like more like a filtering in time , rather than doing a r	Ah. OK. So what i what about , um i u what i w I mean , I don't know if this is a good idea or not , but what if you put ran the other kind of LDA , uh , on your features right before they go into the HMM ?	Uh , it	false	QMSum_194
So this is a two dimensional tile. And the LDA that we are f applying is only in time , not in frequency high cost frequency. So it 's like more like a filtering in time , rather than doing a r	Ah. OK. So what i what about , um i u what i w I mean , I don't know if this is a good idea or not , but what if you put ran the other kind of LDA , uh , on your features right before they go into the HMM ?	Uh , it	Mm - hmm. No , actually , I think i	false	QMSum_194
Ah. OK. So what i what about , um i u what i w I mean , I don't know if this is a good idea or not , but what if you put ran the other kind of LDA , uh , on your features right before they go into the HMM ?	Uh , it	Mm - hmm. No , actually , I think i	Well. What do we do with the ANN is is something like that except that it 's not linear. But it 's it 's like a nonlinear discriminant analysis.	false	QMSum_194
Uh , it	Mm - hmm. No , actually , I think i	Well. What do we do with the ANN is is something like that except that it 's not linear. But it 's it 's like a nonlinear discriminant analysis.	Yeah. Right , it 's the It 's Right. The So Yeah , so it 's sort of like	false	QMSum_194
Mm - hmm. No , actually , I think i	Well. What do we do with the ANN is is something like that except that it 's not linear. But it 's it 's like a nonlinear discriminant analysis.	Yeah. Right , it 's the It 's Right. The So Yeah , so it 's sort of like	But.	false	QMSum_194
Well. What do we do with the ANN is is something like that except that it 's not linear. But it 's it 's like a nonlinear discriminant analysis.	Yeah. Right , it 's the It 's Right. The So Yeah , so it 's sort of like	But.	The tandem stuff is kind of like i nonlinear LDA.	false	QMSum_194
Yeah. Right , it 's the It 's Right. The So Yeah , so it 's sort of like	But.	The tandem stuff is kind of like i nonlinear LDA.	Yeah. It 's	false	QMSum_194
But.	The tandem stuff is kind of like i nonlinear LDA.	Yeah. It 's	I g	false	QMSum_194
The tandem stuff is kind of like i nonlinear LDA.	Yeah. It 's	I g	Yeah.	false	QMSum_194
Yeah. It 's	I g	Yeah.	Yeah.	false	QMSum_194
I g	Yeah.	Yeah.	Yeah.	false	QMSum_194
Yeah.	Yeah.	Yeah.	But I mean , w but the other features that you have , um , th the non - tandem ones ,	false	QMSum_194
Yeah.	Yeah.	But I mean , w but the other features that you have , um , th the non - tandem ones ,	Uh. Mm - hmm. Yeah , I know. That that Yeah. Well , in the proposal , they were transformed u using PCA , but	false	QMSum_194
Yeah.	But I mean , w but the other features that you have , um , th the non - tandem ones ,	Uh. Mm - hmm. Yeah , I know. That that Yeah. Well , in the proposal , they were transformed u using PCA , but	Uh - huh.	false	QMSum_194
But I mean , w but the other features that you have , um , th the non - tandem ones ,	Uh. Mm - hmm. Yeah , I know. That that Yeah. Well , in the proposal , they were transformed u using PCA , but	Uh - huh.	Yeah , it might be that LDA could be better.	false	QMSum_194
Uh. Mm - hmm. Yeah , I know. That that Yeah. Well , in the proposal , they were transformed u using PCA , but	Uh - huh.	Yeah , it might be that LDA could be better.	The a the argument i is kind of i in and it 's not like we really know , but the argument anyway is that , um , uh , we always have the prob I mean , discriminative things are good. LDA , neural nets , they 're good.	false	QMSum_194
Uh - huh.	Yeah , it might be that LDA could be better.	The a the argument i is kind of i in and it 's not like we really know , but the argument anyway is that , um , uh , we always have the prob I mean , discriminative things are good. LDA , neural nets , they 're good.	Yeah.	false	QMSum_194
Yeah , it might be that LDA could be better.	The a the argument i is kind of i in and it 's not like we really know , but the argument anyway is that , um , uh , we always have the prob I mean , discriminative things are good. LDA , neural nets , they 're good.	Yeah.	Uh , they 're good because you you you learn to distinguish between these categories that you want to be good at distinguishing between. And PCA doesn't do that. It PAC - PCA low - order PCA throws away pieces that are uh , maybe not not gonna be helpful just because they 're small , basically.	false	QMSum_194
The a the argument i is kind of i in and it 's not like we really know , but the argument anyway is that , um , uh , we always have the prob I mean , discriminative things are good. LDA , neural nets , they 're good.	Yeah.	Uh , they 're good because you you you learn to distinguish between these categories that you want to be good at distinguishing between. And PCA doesn't do that. It PAC - PCA low - order PCA throws away pieces that are uh , maybe not not gonna be helpful just because they 're small , basically.	Right.	false	QMSum_194
Yeah.	Uh , they 're good because you you you learn to distinguish between these categories that you want to be good at distinguishing between. And PCA doesn't do that. It PAC - PCA low - order PCA throws away pieces that are uh , maybe not not gonna be helpful just because they 're small , basically.	Right.	But , uh , the problem is , training sets aren't perfect and testing sets are different. So you f you you face the potential problem with discriminative stuff , be it LDA or neural nets , that you are training to discriminate between categories in one space but what you 're really gonna be g getting is is something else.	false	QMSum_194
Uh , they 're good because you you you learn to distinguish between these categories that you want to be good at distinguishing between. And PCA doesn't do that. It PAC - PCA low - order PCA throws away pieces that are uh , maybe not not gonna be helpful just because they 're small , basically.	Right.	But , uh , the problem is , training sets aren't perfect and testing sets are different. So you f you you face the potential problem with discriminative stuff , be it LDA or neural nets , that you are training to discriminate between categories in one space but what you 're really gonna be g getting is is something else.	Uh - huh.	false	QMSum_194
Right.	But , uh , the problem is , training sets aren't perfect and testing sets are different. So you f you you face the potential problem with discriminative stuff , be it LDA or neural nets , that you are training to discriminate between categories in one space but what you 're really gonna be g getting is is something else.	Uh - huh.	And so , uh , Stephane 's idea was , uh , let 's feed , uh , both this discriminatively trained thing and something that 's not. So you have a good set of features that everybody 's worked really hard to make ,	false	QMSum_194
But , uh , the problem is , training sets aren't perfect and testing sets are different. So you f you you face the potential problem with discriminative stuff , be it LDA or neural nets , that you are training to discriminate between categories in one space but what you 're really gonna be g getting is is something else.	Uh - huh.	And so , uh , Stephane 's idea was , uh , let 's feed , uh , both this discriminatively trained thing and something that 's not. So you have a good set of features that everybody 's worked really hard to make ,	Yeah.	false	QMSum_194
Uh - huh.	And so , uh , Stephane 's idea was , uh , let 's feed , uh , both this discriminatively trained thing and something that 's not. So you have a good set of features that everybody 's worked really hard to make ,	Yeah.	and then , uh , you you discriminately train it , but you also take the path that that doesn't have that ,	false	QMSum_194
And so , uh , Stephane 's idea was , uh , let 's feed , uh , both this discriminatively trained thing and something that 's not. So you have a good set of features that everybody 's worked really hard to make ,	Yeah.	and then , uh , you you discriminately train it , but you also take the path that that doesn't have that ,	Uh - huh.	false	QMSum_194
Yeah.	and then , uh , you you discriminately train it , but you also take the path that that doesn't have that ,	Uh - huh.	and putting those in together. And that that seem So it 's kind of like a combination of the uh , what , uh , Dan has been calling , you know , a feature uh , you know , a feature combination versus posterior combination or something. It 's it 's , you know , you have the posterior combination but then you get the features from that and use them as a feature combination with these these other things. And that seemed , at least in the last one , as he was just saying , he he when he only did discriminative stuff , i it actually was was it didn't help at all in this particular case.	false	QMSum_194
and then , uh , you you discriminately train it , but you also take the path that that doesn't have that ,	Uh - huh.	and putting those in together. And that that seem So it 's kind of like a combination of the uh , what , uh , Dan has been calling , you know , a feature uh , you know , a feature combination versus posterior combination or something. It 's it 's , you know , you have the posterior combination but then you get the features from that and use them as a feature combination with these these other things. And that seemed , at least in the last one , as he was just saying , he he when he only did discriminative stuff , i it actually was was it didn't help at all in this particular case.	Yeah.	false	QMSum_194
Uh - huh.	and putting those in together. And that that seem So it 's kind of like a combination of the uh , what , uh , Dan has been calling , you know , a feature uh , you know , a feature combination versus posterior combination or something. It 's it 's , you know , you have the posterior combination but then you get the features from that and use them as a feature combination with these these other things. And that seemed , at least in the last one , as he was just saying , he he when he only did discriminative stuff , i it actually was was it didn't help at all in this particular case.	Yeah.	There was enough of a difference , I guess , between the testing and training. But by having them both there The fact is some of the time , the discriminative stuff is gonna help you.	false	QMSum_194
and putting those in together. And that that seem So it 's kind of like a combination of the uh , what , uh , Dan has been calling , you know , a feature uh , you know , a feature combination versus posterior combination or something. It 's it 's , you know , you have the posterior combination but then you get the features from that and use them as a feature combination with these these other things. And that seemed , at least in the last one , as he was just saying , he he when he only did discriminative stuff , i it actually was was it didn't help at all in this particular case.	Yeah.	There was enough of a difference , I guess , between the testing and training. But by having them both there The fact is some of the time , the discriminative stuff is gonna help you.	Mm - hmm.	false	QMSum_194
Yeah.	There was enough of a difference , I guess , between the testing and training. But by having them both there The fact is some of the time , the discriminative stuff is gonna help you.	Mm - hmm.	And some of the time it 's going to hurt you ,	false	QMSum_194
There was enough of a difference , I guess , between the testing and training. But by having them both there The fact is some of the time , the discriminative stuff is gonna help you.	Mm - hmm.	And some of the time it 's going to hurt you ,	Right.	false	QMSum_194

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 ?

Uh , there is a car noise. So there are f just four noises. Um , uh , " Car " , I think , " Babble " ,

false

QMSum_194

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 ?

Uh , there is a car noise. So there are f just four noises. Um , uh , " Car " , I think , " Babble " ,

" Babble. "

false

QMSum_194

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 ?

Uh , there is a car noise. So there are f just four noises. Um , uh , " Car " , I think , " Babble " ,

" Babble. "

" Subway " , right ? and

false

QMSum_194

Uh , there is a car noise. So there are f just four noises. Um , uh , " Car " , I think , " Babble " ,

" Babble. "

" Subway " , right ? and

" Street " or " Airport " or something.

false

QMSum_194

" Babble. "

" Subway " , right ? and

" Street " or " Airport " or something.

and " Street " isn't

false

QMSum_194

" Subway " , right ? and

" Street " or " Airport " or something.

and " Street " isn't

Or " Train station ".

false

QMSum_194

" Street " or " Airport " or something.

and " Street " isn't

Or " Train station ".

" Train station " , yeah.

false

QMSum_194

and " Street " isn't

Or " Train station ".

" Train station " , yeah.

Yeah.

false

QMSum_194

Or " Train station ".

" Train station " , yeah.

Yeah.

So it 's mostly Well , " Car " is stationary ,

false

QMSum_194

" Train station " , yeah.

Yeah.

So it 's mostly Well , " Car " is stationary ,

Mm - hmm.

false

QMSum_194

Yeah.

So it 's mostly Well , " Car " is stationary ,

Mm - hmm.

" Babble " , it 's a stationary background plus some voices ,

false

QMSum_194

So it 's mostly Well , " Car " is stationary ,

Mm - hmm.

" Babble " , it 's a stationary background plus some voices ,

Mm - hmm.

false

QMSum_194

Mm - hmm.

" Babble " , it 's a stationary background plus some voices ,

Mm - hmm.

some speech over it. And the other two are rather stationary also.

false

QMSum_194

" Babble " , it 's a stationary background plus some voices ,

Mm - hmm.

some speech over it. And the other two are rather stationary also.

Well , I I think that if you run it Actually , you maybe you remember this. When you in in the old experiments when you ran with the neural net only , and didn't have this side path , um , uh , with the the pure features as well , did it make things better to have the neural net ?

false

QMSum_194

Mm - hmm.

some speech over it. And the other two are rather stationary also.

Well , I I think that if you run it Actually , you maybe you remember this. When you in in the old experiments when you ran with the neural net only , and didn't have this side path , um , uh , with the the pure features as well , did it make things better to have the neural net ?

Mm - hmm.

false

QMSum_194

some speech over it. And the other two are rather stationary also.

Well , I I think that if you run it Actually , you maybe you remember this. When you in in the old experiments when you ran with the neural net only , and didn't have this side path , um , uh , with the the pure features as well , did it make things better to have the neural net ?

Mm - hmm.

Was it about the same ? Uh , w i

false

QMSum_194

Well , I I think that if you run it Actually , you maybe you remember this. When you in in the old experiments when you ran with the neural net only , and didn't have this side path , um , uh , with the the pure features as well , did it make things better to have the neural net ?

Mm - hmm.

Was it about the same ? Uh , w i

It was b a little bit worse.

false

QMSum_194

Mm - hmm.

Was it about the same ? Uh , w i

It was b a little bit worse.

Than ?

false

QMSum_194

Was it about the same ? Uh , w i

It was b a little bit worse.

Than ?

Than just the features , yeah.

false

QMSum_194

It was b a little bit worse.

Than ?

Than just the features , yeah.

So , until you put the second path in with the pure features , the neural net wasn't helping at all.

false

QMSum_194

Than ?

Than just the features , yeah.

So , until you put the second path in with the pure features , the neural net wasn't helping at all.

Mm - hmm.

false

QMSum_194

Than just the features , yeah.

So , until you put the second path in with the pure features , the neural net wasn't helping at all.

Mm - hmm.

Well , that 's interesting.

false

QMSum_194

So , until you put the second path in with the pure features , the neural net wasn't helping at all.

Mm - hmm.

Well , that 's interesting.

It was helping , uh , if the features are b were bad ,

false

QMSum_194

Mm - hmm.

Well , that 's interesting.

It was helping , uh , if the features are b were bad ,

Yeah.

false

QMSum_194

Well , that 's interesting.

It was helping , uh , if the features are b were bad ,

Yeah.

I mean. Just plain P L Ps or M F

false

QMSum_194

It was helping , uh , if the features are b were bad ,

Yeah.

I mean. Just plain P L Ps or M F

Yeah.

false

QMSum_194

Yeah.

I mean. Just plain P L Ps or M F

Yeah.

C Cs. as soon as we added LDA on - line normalization , and all these things , then

false

QMSum_194

I mean. Just plain P L Ps or M F

Yeah.

C Cs. as soon as we added LDA on - line normalization , and all these things , then

They were doing similar enough things. Well , I still think it would be k sort of interesting to see what would happen if you just had the neural net without the side thing.

false

QMSum_194

Yeah.

C Cs. as soon as we added LDA on - line normalization , and all these things , then

They were doing similar enough things. Well , I still think it would be k sort of interesting to see what would happen if you just had the neural net without the side thing.

Yeah ,

false

QMSum_194

C Cs. as soon as we added LDA on - line normalization , and all these things , then

They were doing similar enough things. Well , I still think it would be k sort of interesting to see what would happen if you just had the neural net without the side thing.

Yeah ,

And and the thing I I have in mind is , uh , maybe you 'll see that the results are not just a little bit worse.

false

QMSum_194

They were doing similar enough things. Well , I still think it would be k sort of interesting to see what would happen if you just had the neural net without the side thing.

Yeah ,

And and the thing I I have in mind is , uh , maybe you 'll see that the results are not just a little bit worse.

mm - hmm.

false

QMSum_194

Yeah ,

And and the thing I I have in mind is , uh , maybe you 'll see that the results are not just a little bit worse.

mm - hmm.

Maybe that they 're a lot worse. You know ? And , um But if on the ha other hand , uh , it 's , say , somewhere in between what you 're seeing now and and and , uh , what you 'd have with just the pure features , then maybe there is some problem of a of a , uh , combination of these things , or correlation between them somehow.

false

QMSum_194

And and the thing I I have in mind is , uh , maybe you 'll see that the results are not just a little bit worse.

mm - hmm.

Maybe that they 're a lot worse. You know ? And , um But if on the ha other hand , uh , it 's , say , somewhere in between what you 're seeing now and and and , uh , what you 'd have with just the pure features , then maybe there is some problem of a of a , uh , combination of these things , or correlation between them somehow.

Mm - hmm.

false

QMSum_194

mm - hmm.

Maybe that they 're a lot worse. You know ? And , um But if on the ha other hand , uh , it 's , say , somewhere in between what you 're seeing now and and and , uh , what you 'd have with just the pure features , then maybe there is some problem of a of a , uh , combination of these things , or correlation between them somehow.

Mm - hmm.

If it really is that the net is hurting you at the moment , then I think the issue is to focus on on , uh , improving the the net.

false

QMSum_194

Maybe that they 're a lot worse. You know ? And , um But if on the ha other hand , uh , it 's , say , somewhere in between what you 're seeing now and and and , uh , what you 'd have with just the pure features , then maybe there is some problem of a of a , uh , combination of these things , or correlation between them somehow.

Mm - hmm.

If it really is that the net is hurting you at the moment , then I think the issue is to focus on on , uh , improving the the net.

Yeah ,

false

QMSum_194

Mm - hmm.

If it really is that the net is hurting you at the moment , then I think the issue is to focus on on , uh , improving the the net.

Yeah ,

Um.

false

QMSum_194

If it really is that the net is hurting you at the moment , then I think the issue is to focus on on , uh , improving the the net.

Yeah ,

Um.

mm - hmm.

false

QMSum_194

Yeah ,

Um.

mm - hmm.

So what 's the overall effe I mean , you haven't done all the experiments but you said it was i somewhat better , say , five percent better , for the first two conditions , and fifteen percent worse for the other one ? But it 's but of course that one 's weighted lower ,

false

QMSum_194

Um.

mm - hmm.

So what 's the overall effe I mean , you haven't done all the experiments but you said it was i somewhat better , say , five percent better , for the first two conditions , and fifteen percent worse for the other one ? But it 's but of course that one 's weighted lower ,

Y yeah , oh. Yeah.

false

QMSum_194

mm - hmm.

So what 's the overall effe I mean , you haven't done all the experiments but you said it was i somewhat better , say , five percent better , for the first two conditions , and fifteen percent worse for the other one ? But it 's but of course that one 's weighted lower ,

Y yeah , oh. Yeah.

so I wonder what the net effect is.

false

QMSum_194

So what 's the overall effe I mean , you haven't done all the experiments but you said it was i somewhat better , say , five percent better , for the first two conditions , and fifteen percent worse for the other one ? But it 's but of course that one 's weighted lower ,

Y yeah , oh. Yeah.

so I wonder what the net effect is.

I d I I think it 's it was one or two percent. That 's not that bad , but it was l like two percent relative worse on SpeechDat - Car. I have to to check that. Well , I have I will.

false

QMSum_194

Y yeah , oh. Yeah.

so I wonder what the net effect is.

I d I I think it 's it was one or two percent. That 's not that bad , but it was l like two percent relative worse on SpeechDat - Car. I have to to check that. Well , I have I will.

Well , it will overall it will be still better even if it is fifteen percent worse , because the fifteen percent worse is given like f w twenty - five point two five eight.

false

QMSum_194

so I wonder what the net effect is.

I d I I think it 's it was one or two percent. That 's not that bad , but it was l like two percent relative worse on SpeechDat - Car. I have to to check that. Well , I have I will.

Well , it will overall it will be still better even if it is fifteen percent worse , because the fifteen percent worse is given like f w twenty - five point two five eight.

Right.

false

QMSum_194

I d I I think it 's it was one or two percent. That 's not that bad , but it was l like two percent relative worse on SpeechDat - Car. I have to to check that. Well , I have I will.

Well , it will overall it will be still better even if it is fifteen percent worse , because the fifteen percent worse is given like f w twenty - five point two five eight.

Right.

Mm - hmm. Hmm.

false

QMSum_194

Well , it will overall it will be still better even if it is fifteen percent worse , because the fifteen percent worse is given like f w twenty - five point two five eight.

Right.

Mm - hmm. Hmm.

Right. So the so the worst it could be , if the others were exactly the same , is four ,

false

QMSum_194

Right.

Mm - hmm. Hmm.

Right. So the so the worst it could be , if the others were exactly the same , is four ,

Is it like

false

QMSum_194

Mm - hmm. Hmm.

Right. So the so the worst it could be , if the others were exactly the same , is four ,

Is it like

and and , uh , in fact since the others are somewhat better

false

QMSum_194

Right. So the so the worst it could be , if the others were exactly the same , is four ,

Is it like

and and , uh , in fact since the others are somewhat better

Yeah , so it 's four. Is i So either it 'll get cancelled out , or you 'll get , like , almost the same.

false

QMSum_194

Is it like

and and , uh , in fact since the others are somewhat better

Yeah , so it 's four. Is i So either it 'll get cancelled out , or you 'll get , like , almost the same.

Uh.

false

QMSum_194

and and , uh , in fact since the others are somewhat better

Yeah , so it 's four. Is i So either it 'll get cancelled out , or you 'll get , like , almost the same.

Uh.

Yeah , it was it was slightly worse.

false

QMSum_194

Yeah , so it 's four. Is i So either it 'll get cancelled out , or you 'll get , like , almost the same.

Uh.

Yeah , it was it was slightly worse.

Slightly bad. Yeah.

false

QMSum_194

Uh.

Yeah , it was it was slightly worse.

Slightly bad. Yeah.

Um ,

false

QMSum_194

Yeah , it was it was slightly worse.

Slightly bad. Yeah.

Um ,

Yeah , it should be pretty close to cancelled out.

false

QMSum_194

Slightly bad. Yeah.

Um ,

Yeah , it should be pretty close to cancelled out.

Yeah.

false

QMSum_194

Um ,

Yeah , it should be pretty close to cancelled out.

Yeah.

You know , I 've been wondering about something.

false

QMSum_194

Yeah , it should be pretty close to cancelled out.

Yeah.

You know , I 've been wondering about something.

Mm - hmm.

false

QMSum_194

Yeah.

You know , I 've been wondering about something.

Mm - hmm.

In the , um a lot of the , um the Hub - five systems , um , recently have been using LDA. and and they , um They run LDA on the features right before they train the models. So there 's the the LDA is is right there before the H M

false

QMSum_194

You know , I 've been wondering about something.

Mm - hmm.

In the , um a lot of the , um the Hub - five systems , um , recently have been using LDA. and and they , um They run LDA on the features right before they train the models. So there 's the the LDA is is right there before the H M

Yeah.

false

QMSum_194

Mm - hmm.

In the , um a lot of the , um the Hub - five systems , um , recently have been using LDA. and and they , um They run LDA on the features right before they train the models. So there 's the the LDA is is right there before the H M

Yeah.

So , you guys are using LDA but it seems like it 's pretty far back in the process.

false

QMSum_194

In the , um a lot of the , um the Hub - five systems , um , recently have been using LDA. and and they , um They run LDA on the features right before they train the models. So there 's the the LDA is is right there before the H M

Yeah.

So , you guys are using LDA but it seems like it 's pretty far back in the process.

Uh , this LDA is different from the LDA that you are talking about. The LDA that you saying is , like , you take a block of features , like nine frames or something , and then do an LDA on it ,

false

QMSum_194

Yeah.

So , you guys are using LDA but it seems like it 's pretty far back in the process.

Uh , this LDA is different from the LDA that you are talking about. The LDA that you saying is , like , you take a block of features , like nine frames or something , and then do an LDA on it ,

Yeah. Uh - huh.

false

QMSum_194

So , you guys are using LDA but it seems like it 's pretty far back in the process.

Uh , this LDA is different from the LDA that you are talking about. The LDA that you saying is , like , you take a block of features , like nine frames or something , and then do an LDA on it ,

Yeah. Uh - huh.

and then reduce the dimensionality to something like twenty - four or something like that.

false

QMSum_194

Uh , this LDA is different from the LDA that you are talking about. The LDA that you saying is , like , you take a block of features , like nine frames or something , and then do an LDA on it ,

Yeah. Uh - huh.

and then reduce the dimensionality to something like twenty - four or something like that.

Yeah , you c you c you can.

false

QMSum_194

Yeah. Uh - huh.

and then reduce the dimensionality to something like twenty - four or something like that.

Yeah , you c you c you can.

And then feed it to HMM.

false

QMSum_194

and then reduce the dimensionality to something like twenty - four or something like that.

Yeah , you c you c you can.

And then feed it to HMM.

I mean , it 's you know , you 're just basically i

false

QMSum_194

Yeah , you c you c you can.

And then feed it to HMM.

I mean , it 's you know , you 're just basically i

Yeah , so this is like a two d two dimensional tile.

false

QMSum_194

And then feed it to HMM.

I mean , it 's you know , you 're just basically i

Yeah , so this is like a two d two dimensional tile.

You 're shifting the feature space. Yeah.

false

QMSum_194

I mean , it 's you know , you 're just basically i

Yeah , so this is like a two d two dimensional tile.

You 're shifting the feature space. Yeah.

So this is a two dimensional tile. And the LDA that we are f applying is only in time , not in frequency high cost frequency. So it 's like more like a filtering in time , rather than doing a r

false

QMSum_194

Yeah , so this is like a two d two dimensional tile.

You 're shifting the feature space. Yeah.

So this is a two dimensional tile. And the LDA that we are f applying is only in time , not in frequency high cost frequency. So it 's like more like a filtering in time , rather than doing a r

Ah. OK. So what i what about , um i u what i w I mean , I don't know if this is a good idea or not , but what if you put ran the other kind of LDA , uh , on your features right before they go into the HMM ?

false

QMSum_194

You 're shifting the feature space. Yeah.

So this is a two dimensional tile. And the LDA that we are f applying is only in time , not in frequency high cost frequency. So it 's like more like a filtering in time , rather than doing a r

Ah. OK. So what i what about , um i u what i w I mean , I don't know if this is a good idea or not , but what if you put ran the other kind of LDA , uh , on your features right before they go into the HMM ?

Uh , it

false

QMSum_194

So this is a two dimensional tile. And the LDA that we are f applying is only in time , not in frequency high cost frequency. So it 's like more like a filtering in time , rather than doing a r

Ah. OK. So what i what about , um i u what i w I mean , I don't know if this is a good idea or not , but what if you put ran the other kind of LDA , uh , on your features right before they go into the HMM ?

Uh , it

Mm - hmm. No , actually , I think i

false

QMSum_194

Ah. OK. So what i what about , um i u what i w I mean , I don't know if this is a good idea or not , but what if you put ran the other kind of LDA , uh , on your features right before they go into the HMM ?

Uh , it

Mm - hmm. No , actually , I think i

Well. What do we do with the ANN is is something like that except that it 's not linear. But it 's it 's like a nonlinear discriminant analysis.

false

QMSum_194

Uh , it

Mm - hmm. No , actually , I think i

Well. What do we do with the ANN is is something like that except that it 's not linear. But it 's it 's like a nonlinear discriminant analysis.

Yeah. Right , it 's the It 's Right. The So Yeah , so it 's sort of like

false

QMSum_194

Mm - hmm. No , actually , I think i

Well. What do we do with the ANN is is something like that except that it 's not linear. But it 's it 's like a nonlinear discriminant analysis.

Yeah. Right , it 's the It 's Right. The So Yeah , so it 's sort of like

But.

false

QMSum_194

Well. What do we do with the ANN is is something like that except that it 's not linear. But it 's it 's like a nonlinear discriminant analysis.

Yeah. Right , it 's the It 's Right. The So Yeah , so it 's sort of like

But.

The tandem stuff is kind of like i nonlinear LDA.

false

QMSum_194

Yeah. Right , it 's the It 's Right. The So Yeah , so it 's sort of like

But.

The tandem stuff is kind of like i nonlinear LDA.

Yeah. It 's

false

QMSum_194

But.

The tandem stuff is kind of like i nonlinear LDA.

Yeah. It 's

I g

false

QMSum_194

The tandem stuff is kind of like i nonlinear LDA.

Yeah. It 's

I g

Yeah.

false

QMSum_194

Yeah. It 's

I g

Yeah.

false

QMSum_194

I g

Yeah.

false

QMSum_194

Yeah.

But I mean , w but the other features that you have , um , th the non - tandem ones ,

false

QMSum_194

Yeah.

But I mean , w but the other features that you have , um , th the non - tandem ones ,

Uh. Mm - hmm. Yeah , I know. That that Yeah. Well , in the proposal , they were transformed u using PCA , but

false

QMSum_194

Yeah.

But I mean , w but the other features that you have , um , th the non - tandem ones ,

Uh. Mm - hmm. Yeah , I know. That that Yeah. Well , in the proposal , they were transformed u using PCA , but

Uh - huh.

false

QMSum_194

But I mean , w but the other features that you have , um , th the non - tandem ones ,

Uh. Mm - hmm. Yeah , I know. That that Yeah. Well , in the proposal , they were transformed u using PCA , but

Uh - huh.

Yeah , it might be that LDA could be better.

false

QMSum_194

Uh. Mm - hmm. Yeah , I know. That that Yeah. Well , in the proposal , they were transformed u using PCA , but

Uh - huh.

Yeah , it might be that LDA could be better.

The a the argument i is kind of i in and it 's not like we really know , but the argument anyway is that , um , uh , we always have the prob I mean , discriminative things are good. LDA , neural nets , they 're good.

false

QMSum_194

Uh - huh.

Yeah , it might be that LDA could be better.

The a the argument i is kind of i in and it 's not like we really know , but the argument anyway is that , um , uh , we always have the prob I mean , discriminative things are good. LDA , neural nets , they 're good.

Yeah.

false

QMSum_194

Yeah , it might be that LDA could be better.

The a the argument i is kind of i in and it 's not like we really know , but the argument anyway is that , um , uh , we always have the prob I mean , discriminative things are good. LDA , neural nets , they 're good.

Yeah.

Uh , they 're good because you you you learn to distinguish between these categories that you want to be good at distinguishing between. And PCA doesn't do that. It PAC - PCA low - order PCA throws away pieces that are uh , maybe not not gonna be helpful just because they 're small , basically.

false

QMSum_194

The a the argument i is kind of i in and it 's not like we really know , but the argument anyway is that , um , uh , we always have the prob I mean , discriminative things are good. LDA , neural nets , they 're good.

Yeah.

Uh , they 're good because you you you learn to distinguish between these categories that you want to be good at distinguishing between. And PCA doesn't do that. It PAC - PCA low - order PCA throws away pieces that are uh , maybe not not gonna be helpful just because they 're small , basically.

Right.

false

QMSum_194

Yeah.

Uh , they 're good because you you you learn to distinguish between these categories that you want to be good at distinguishing between. And PCA doesn't do that. It PAC - PCA low - order PCA throws away pieces that are uh , maybe not not gonna be helpful just because they 're small , basically.

Right.

But , uh , the problem is , training sets aren't perfect and testing sets are different. So you f you you face the potential problem with discriminative stuff , be it LDA or neural nets , that you are training to discriminate between categories in one space but what you 're really gonna be g getting is is something else.

false

QMSum_194

Uh , they 're good because you you you learn to distinguish between these categories that you want to be good at distinguishing between. And PCA doesn't do that. It PAC - PCA low - order PCA throws away pieces that are uh , maybe not not gonna be helpful just because they 're small , basically.

Right.

But , uh , the problem is , training sets aren't perfect and testing sets are different. So you f you you face the potential problem with discriminative stuff , be it LDA or neural nets , that you are training to discriminate between categories in one space but what you 're really gonna be g getting is is something else.

Uh - huh.

false

QMSum_194

Right.

But , uh , the problem is , training sets aren't perfect and testing sets are different. So you f you you face the potential problem with discriminative stuff , be it LDA or neural nets , that you are training to discriminate between categories in one space but what you 're really gonna be g getting is is something else.

Uh - huh.

And so , uh , Stephane 's idea was , uh , let 's feed , uh , both this discriminatively trained thing and something that 's not. So you have a good set of features that everybody 's worked really hard to make ,

false

QMSum_194

But , uh , the problem is , training sets aren't perfect and testing sets are different. So you f you you face the potential problem with discriminative stuff , be it LDA or neural nets , that you are training to discriminate between categories in one space but what you 're really gonna be g getting is is something else.

Uh - huh.

And so , uh , Stephane 's idea was , uh , let 's feed , uh , both this discriminatively trained thing and something that 's not. So you have a good set of features that everybody 's worked really hard to make ,

Yeah.

false

QMSum_194

Uh - huh.

And so , uh , Stephane 's idea was , uh , let 's feed , uh , both this discriminatively trained thing and something that 's not. So you have a good set of features that everybody 's worked really hard to make ,

Yeah.

and then , uh , you you discriminately train it , but you also take the path that that doesn't have that ,

false

QMSum_194

And so , uh , Stephane 's idea was , uh , let 's feed , uh , both this discriminatively trained thing and something that 's not. So you have a good set of features that everybody 's worked really hard to make ,

Yeah.

and then , uh , you you discriminately train it , but you also take the path that that doesn't have that ,

Uh - huh.

false

QMSum_194

Yeah.

and then , uh , you you discriminately train it , but you also take the path that that doesn't have that ,

Uh - huh.

and putting those in together. And that that seem So it 's kind of like a combination of the uh , what , uh , Dan has been calling , you know , a feature uh , you know , a feature combination versus posterior combination or something. It 's it 's , you know , you have the posterior combination but then you get the features from that and use them as a feature combination with these these other things. And that seemed , at least in the last one , as he was just saying , he he when he only did discriminative stuff , i it actually was was it didn't help at all in this particular case.

false

QMSum_194

and then , uh , you you discriminately train it , but you also take the path that that doesn't have that ,

Uh - huh.

and putting those in together. And that that seem So it 's kind of like a combination of the uh , what , uh , Dan has been calling , you know , a feature uh , you know , a feature combination versus posterior combination or something. It 's it 's , you know , you have the posterior combination but then you get the features from that and use them as a feature combination with these these other things. And that seemed , at least in the last one , as he was just saying , he he when he only did discriminative stuff , i it actually was was it didn't help at all in this particular case.

Yeah.

false

QMSum_194

Uh - huh.

and putting those in together. And that that seem So it 's kind of like a combination of the uh , what , uh , Dan has been calling , you know , a feature uh , you know , a feature combination versus posterior combination or something. It 's it 's , you know , you have the posterior combination but then you get the features from that and use them as a feature combination with these these other things. And that seemed , at least in the last one , as he was just saying , he he when he only did discriminative stuff , i it actually was was it didn't help at all in this particular case.

Yeah.

There was enough of a difference , I guess , between the testing and training. But by having them both there The fact is some of the time , the discriminative stuff is gonna help you.

false

QMSum_194

and putting those in together. And that that seem So it 's kind of like a combination of the uh , what , uh , Dan has been calling , you know , a feature uh , you know , a feature combination versus posterior combination or something. It 's it 's , you know , you have the posterior combination but then you get the features from that and use them as a feature combination with these these other things. And that seemed , at least in the last one , as he was just saying , he he when he only did discriminative stuff , i it actually was was it didn't help at all in this particular case.

Yeah.

There was enough of a difference , I guess , between the testing and training. But by having them both there The fact is some of the time , the discriminative stuff is gonna help you.

Mm - hmm.

false

QMSum_194

Yeah.

There was enough of a difference , I guess , between the testing and training. But by having them both there The fact is some of the time , the discriminative stuff is gonna help you.

Mm - hmm.

And some of the time it 's going to hurt you ,

false

QMSum_194

There was enough of a difference , I guess , between the testing and training. But by having them both there The fact is some of the time , the discriminative stuff is gonna help you.

Mm - hmm.

And some of the time it 's going to hurt you ,

Right.

false

QMSum_194