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
and	what sorts of features are you looking at ?	We have some	So we would be looking at , um , the variance of the spectrum of the excitation ,	false	QMSum_120
what sorts of features are you looking at ?	We have some	So we would be looking at , um , the variance of the spectrum of the excitation ,	uh , um , this , this , and this.	false	QMSum_120
We have some	So we would be looking at , um , the variance of the spectrum of the excitation ,	uh , um , this , this , and this.	something like this , which is should be high for voiced sounds. Uh , we	false	QMSum_120
So we would be looking at , um , the variance of the spectrum of the excitation ,	uh , um , this , this , and this.	something like this , which is should be high for voiced sounds. Uh , we	Wait a minute. I what does that mean ? The variance of the spectrum of excitation.	false	QMSum_120
uh , um , this , this , and this.	something like this , which is should be high for voiced sounds. Uh , we	Wait a minute. I what does that mean ? The variance of the spectrum of excitation.	Yeah. So the So basically the spectrum of the excitation for a purely periodic sig signal shou sh	false	QMSum_120
something like this , which is should be high for voiced sounds. Uh , we	Wait a minute. I what does that mean ? The variance of the spectrum of excitation.	Yeah. So the So basically the spectrum of the excitation for a purely periodic sig signal shou sh	OK. Yeah , w what yo what you 're calling the excitation , as I recall , is you 're subtracting the the , um the mel mel mel filter , uh , spectrum from the FFT spectrum.	false	QMSum_120
Wait a minute. I what does that mean ? The variance of the spectrum of excitation.	Yeah. So the So basically the spectrum of the excitation for a purely periodic sig signal shou sh	OK. Yeah , w what yo what you 're calling the excitation , as I recall , is you 're subtracting the the , um the mel mel mel filter , uh , spectrum from the FFT spectrum.	e That 's right. Yeah. So	false	QMSum_120
Yeah. So the So basically the spectrum of the excitation for a purely periodic sig signal shou sh	OK. Yeah , w what yo what you 're calling the excitation , as I recall , is you 're subtracting the the , um the mel mel mel filter , uh , spectrum from the FFT spectrum.	e That 's right. Yeah. So	Right.	false	QMSum_120
OK. Yeah , w what yo what you 're calling the excitation , as I recall , is you 're subtracting the the , um the mel mel mel filter , uh , spectrum from the FFT spectrum.	e That 's right. Yeah. So	Right.	Yeah.	false	QMSum_120
e That 's right. Yeah. So	Right.	Yeah.	Mm - hmm.	false	QMSum_120
Right.	Yeah.	Mm - hmm.	So we have the mel f filter bank , we have the FFT , so we just	false	QMSum_120
Yeah.	Mm - hmm.	So we have the mel f filter bank , we have the FFT , so we just	So it 's it 's not really an excitation ,	false	QMSum_120
Mm - hmm.	So we have the mel f filter bank , we have the FFT , so we just	So it 's it 's not really an excitation ,	No.	false	QMSum_120
So we have the mel f filter bank , we have the FFT , so we just	So it 's it 's not really an excitation ,	No.	but it 's something that hopefully tells you something about the excitation.	false	QMSum_120
So it 's it 's not really an excitation ,	No.	but it 's something that hopefully tells you something about the excitation.	Yeah , that 's right.	false	QMSum_120
No.	but it 's something that hopefully tells you something about the excitation.	Yeah , that 's right.	Yeah , yeah.	false	QMSum_120
but it 's something that hopefully tells you something about the excitation.	Yeah , that 's right.	Yeah , yeah.	Um Yeah.	false	QMSum_120
Yeah , that 's right.	Yeah , yeah.	Um Yeah.	We have here some histogram ,	false	QMSum_120
Yeah , yeah.	Um Yeah.	We have here some histogram ,	E yeah ,	false	QMSum_120
Um Yeah.	We have here some histogram ,	E yeah ,	but they have a lot of overlap.	false	QMSum_120
We have here some histogram ,	E yeah ,	but they have a lot of overlap.	but it 's it 's still Yeah. So , well , for unvoiced portion we have something tha that has a mean around O point three , and for voiced portion the mean is O point fifty - nine. But the variance seem quite high.	false	QMSum_120
E yeah ,	but they have a lot of overlap.	but it 's it 's still Yeah. So , well , for unvoiced portion we have something tha that has a mean around O point three , and for voiced portion the mean is O point fifty - nine. But the variance seem quite high.	How do you know ?	false	QMSum_120
but they have a lot of overlap.	but it 's it 's still Yeah. So , well , for unvoiced portion we have something tha that has a mean around O point three , and for voiced portion the mean is O point fifty - nine. But the variance seem quite high.	How do you know ?	So Mmm.	false	QMSum_120
but it 's it 's still Yeah. So , well , for unvoiced portion we have something tha that has a mean around O point three , and for voiced portion the mean is O point fifty - nine. But the variance seem quite high.	How do you know ?	So Mmm.	How did you get your voiced and unvoiced truth data ?	false	QMSum_120
How do you know ?	So Mmm.	How did you get your voiced and unvoiced truth data ?	We used , uh , TIMIT and we used canonical mappings between the phones	false	QMSum_120
So Mmm.	How did you get your voiced and unvoiced truth data ?	We used , uh , TIMIT and we used canonical mappings between the phones	Yeah. We , uh , use TIMIT on this ,	false	QMSum_120
How did you get your voiced and unvoiced truth data ?	We used , uh , TIMIT and we used canonical mappings between the phones	Yeah. We , uh , use TIMIT on this ,	and	false	QMSum_120
We used , uh , TIMIT and we used canonical mappings between the phones	Yeah. We , uh , use TIMIT on this ,	and	for	false	QMSum_120
Yeah. We , uh , use TIMIT on this ,	and	for	th Yeah.	false	QMSum_120
and	for	th Yeah.	But if we look at it in one sentence , it apparently it 's good , I think.	false	QMSum_120
for	th Yeah.	But if we look at it in one sentence , it apparently it 's good , I think.	Yeah , but Yeah. Uh , so it 's noisy TIMIT. That 's right. Yeah.	false	QMSum_120
th Yeah.	But if we look at it in one sentence , it apparently it 's good , I think.	Yeah , but Yeah. Uh , so it 's noisy TIMIT. That 's right. Yeah.	It 's noisy TIMIT.	false	QMSum_120
But if we look at it in one sentence , it apparently it 's good , I think.	Yeah , but Yeah. Uh , so it 's noisy TIMIT. That 's right. Yeah.	It 's noisy TIMIT.	Yeah.	false	QMSum_120
Yeah , but Yeah. Uh , so it 's noisy TIMIT. That 's right. Yeah.	It 's noisy TIMIT.	Yeah.	It seems quite robust to noise , so when we take we draw its parameters across time for a clean sentence and then nois the same noisy sentence , it 's very close.	false	QMSum_120
It 's noisy TIMIT.	Yeah.	It seems quite robust to noise , so when we take we draw its parameters across time for a clean sentence and then nois the same noisy sentence , it 's very close.	Mm - hmm.	false	QMSum_120
Yeah.	It seems quite robust to noise , so when we take we draw its parameters across time for a clean sentence and then nois the same noisy sentence , it 's very close.	Mm - hmm.	Yeah. So there are there is this. There could be also the , um something like the maximum of the auto - correlation function or which	false	QMSum_120
It seems quite robust to noise , so when we take we draw its parameters across time for a clean sentence and then nois the same noisy sentence , it 's very close.	Mm - hmm.	Yeah. So there are there is this. There could be also the , um something like the maximum of the auto - correlation function or which	Is this a a s a trained system ? Or is it a system where you just pick some thresholds ? Ho - how does it work ?	false	QMSum_120
Mm - hmm.	Yeah. So there are there is this. There could be also the , um something like the maximum of the auto - correlation function or which	Is this a a s a trained system ? Or is it a system where you just pick some thresholds ? Ho - how does it work ?	Right now we just are trying to find some features. And ,	false	QMSum_120
Yeah. So there are there is this. There could be also the , um something like the maximum of the auto - correlation function or which	Is this a a s a trained system ? Or is it a system where you just pick some thresholds ? Ho - how does it work ?	Right now we just are trying to find some features. And ,	Mm - hmm.	false	QMSum_120
Is this a a s a trained system ? Or is it a system where you just pick some thresholds ? Ho - how does it work ?	Right now we just are trying to find some features. And ,	Mm - hmm.	uh Yeah. Hopefully , I think what we want to have is to put these features in s some kind of , um well , to to obtain a statistical model on these features and to or just to use a neural network and hopefully these features w would help	false	QMSum_120
Right now we just are trying to find some features. And ,	Mm - hmm.	uh Yeah. Hopefully , I think what we want to have is to put these features in s some kind of , um well , to to obtain a statistical model on these features and to or just to use a neural network and hopefully these features w would help	Because it seems like what you said about the mean of the the voiced and the unvoiced that seemed pretty encouraging.	false	QMSum_120
Mm - hmm.	uh Yeah. Hopefully , I think what we want to have is to put these features in s some kind of , um well , to to obtain a statistical model on these features and to or just to use a neural network and hopefully these features w would help	Because it seems like what you said about the mean of the the voiced and the unvoiced that seemed pretty encouraging.	Mm - hmm.	false	QMSum_120
uh Yeah. Hopefully , I think what we want to have is to put these features in s some kind of , um well , to to obtain a statistical model on these features and to or just to use a neural network and hopefully these features w would help	Because it seems like what you said about the mean of the the voiced and the unvoiced that seemed pretty encouraging.	Mm - hmm.	Well , yeah , except the variance was big.	false	QMSum_120
Because it seems like what you said about the mean of the the voiced and the unvoiced that seemed pretty encouraging.	Mm - hmm.	Well , yeah , except the variance was big.	Right ?	false	QMSum_120
Mm - hmm.	Well , yeah , except the variance was big.	Right ?	Yeah. Except the variance is quite high.	false	QMSum_120
Well , yeah , except the variance was big.	Right ?	Yeah. Except the variance is quite high.	Right ?	false	QMSum_120
Right ?	Yeah. Except the variance is quite high.	Right ?	Well , y	false	QMSum_120
Yeah. Except the variance is quite high.	Right ?	Well , y	Yeah.	false	QMSum_120
Right ?	Well , y	Yeah.	Well , y I I don't know that I would trust that so much because you 're doing these canonical mappings from TIMIT labellings.	false	QMSum_120
Well , y	Yeah.	Well , y I I don't know that I would trust that so much because you 're doing these canonical mappings from TIMIT labellings.	Uh - huh.	false	QMSum_120
Yeah.	Well , y I I don't know that I would trust that so much because you 're doing these canonical mappings from TIMIT labellings.	Uh - huh.	Right ? So , really that 's sort of a cartoon picture about what 's voiced and unvoiced. So that could be giving you a lot of variance.	false	QMSum_120
Well , y I I don't know that I would trust that so much because you 're doing these canonical mappings from TIMIT labellings.	Uh - huh.	Right ? So , really that 's sort of a cartoon picture about what 's voiced and unvoiced. So that could be giving you a lot of variance.	Yeah.	false	QMSum_120
Uh - huh.	Right ? So , really that 's sort of a cartoon picture about what 's voiced and unvoiced. So that could be giving you a lot of variance.	Yeah.	I mean , i it it may be that that you 're finding something good and that the variance is sort of artificial because of how you 're getting your truth.	false	QMSum_120
Right ? So , really that 's sort of a cartoon picture about what 's voiced and unvoiced. So that could be giving you a lot of variance.	Yeah.	I mean , i it it may be that that you 're finding something good and that the variance is sort of artificial because of how you 're getting your truth.	Mm - hmm.	false	QMSum_120
Yeah.	I mean , i it it may be that that you 're finding something good and that the variance is sort of artificial because of how you 're getting your truth.	Mm - hmm.	Yeah. But another way of looking at it might be that I mean , what w we we are coming up with feature sets after all. So another way of looking at it is that um , the mel cepstru mel spectrum , mel cepstrum , any of these variants , um , give you the smooth spectrum. It 's the spectral envelope. By going back to the FFT , you 're getting something that is more like the raw data. So the question is , what characterization and you 're playing around with this another way of looking at it is what characterization of the difference between the raw data and this smooth version is something that you 're missing that could help ? So , I mean , looking at different statistical measures of that difference , coming up with some things and just trying them out and seeing if you add them onto the feature vector does that make things better or worse in noise , where you 're really just i i the way I 'm looking at it is not so much you 're trying to f find the best the world 's best voiced - unvoiced , uh , uh , classifier ,	false	QMSum_120
I mean , i it it may be that that you 're finding something good and that the variance is sort of artificial because of how you 're getting your truth.	Mm - hmm.	Yeah. But another way of looking at it might be that I mean , what w we we are coming up with feature sets after all. So another way of looking at it is that um , the mel cepstru mel spectrum , mel cepstrum , any of these variants , um , give you the smooth spectrum. It 's the spectral envelope. By going back to the FFT , you 're getting something that is more like the raw data. So the question is , what characterization and you 're playing around with this another way of looking at it is what characterization of the difference between the raw data and this smooth version is something that you 're missing that could help ? So , I mean , looking at different statistical measures of that difference , coming up with some things and just trying them out and seeing if you add them onto the feature vector does that make things better or worse in noise , where you 're really just i i the way I 'm looking at it is not so much you 're trying to f find the best the world 's best voiced - unvoiced , uh , uh , classifier ,	Mm - hmm.	false	QMSum_120
Mm - hmm.	Yeah. But another way of looking at it might be that I mean , what w we we are coming up with feature sets after all. So another way of looking at it is that um , the mel cepstru mel spectrum , mel cepstrum , any of these variants , um , give you the smooth spectrum. It 's the spectral envelope. By going back to the FFT , you 're getting something that is more like the raw data. So the question is , what characterization and you 're playing around with this another way of looking at it is what characterization of the difference between the raw data and this smooth version is something that you 're missing that could help ? So , I mean , looking at different statistical measures of that difference , coming up with some things and just trying them out and seeing if you add them onto the feature vector does that make things better or worse in noise , where you 're really just i i the way I 'm looking at it is not so much you 're trying to f find the best the world 's best voiced - unvoiced , uh , uh , classifier ,	Mm - hmm.	Mmm.	false	QMSum_120
Yeah. But another way of looking at it might be that I mean , what w we we are coming up with feature sets after all. So another way of looking at it is that um , the mel cepstru mel spectrum , mel cepstrum , any of these variants , um , give you the smooth spectrum. It 's the spectral envelope. By going back to the FFT , you 're getting something that is more like the raw data. So the question is , what characterization and you 're playing around with this another way of looking at it is what characterization of the difference between the raw data and this smooth version is something that you 're missing that could help ? So , I mean , looking at different statistical measures of that difference , coming up with some things and just trying them out and seeing if you add them onto the feature vector does that make things better or worse in noise , where you 're really just i i the way I 'm looking at it is not so much you 're trying to f find the best the world 's best voiced - unvoiced , uh , uh , classifier ,	Mm - hmm.	Mmm.	but it 's more that , you know , uh , uh , try some different statistical characterizations of that difference back to the raw data	false	QMSum_120
Mm - hmm.	Mmm.	but it 's more that , you know , uh , uh , try some different statistical characterizations of that difference back to the raw data	Right.	false	QMSum_120
Mmm.	but it 's more that , you know , uh , uh , try some different statistical characterizations of that difference back to the raw data	Right.	and and m maybe there 's something there that the system can use.	false	QMSum_120
but it 's more that , you know , uh , uh , try some different statistical characterizations of that difference back to the raw data	Right.	and and m maybe there 's something there that the system can use.	Right.	false	QMSum_120
Right.	and and m maybe there 's something there that the system can use.	Right.	Yeah. Yeah , but ther more obvious is that Yeah. The the more obvious is that that well , using the th the FFT , um , you just it gives you just information about if it 's voiced or not voiced , ma mainly , I mean. But So ,	false	QMSum_120
and and m maybe there 's something there that the system can use.	Right.	Yeah. Yeah , but ther more obvious is that Yeah. The the more obvious is that that well , using the th the FFT , um , you just it gives you just information about if it 's voiced or not voiced , ma mainly , I mean. But So ,	Yeah.	false	QMSum_120
Right.	Yeah. Yeah , but ther more obvious is that Yeah. The the more obvious is that that well , using the th the FFT , um , you just it gives you just information about if it 's voiced or not voiced , ma mainly , I mean. But So ,	Yeah.	this is why we we started to look by having sort of voiced phonemes	false	QMSum_120
Yeah. Yeah , but ther more obvious is that Yeah. The the more obvious is that that well , using the th the FFT , um , you just it gives you just information about if it 's voiced or not voiced , ma mainly , I mean. But So ,	Yeah.	this is why we we started to look by having sort of voiced phonemes	Well , that 's the rea w w what I 'm arguing is that 's Yeah. I mean , uh , what I 'm arguing is that that that 's givi you gives you your intuition.	false	QMSum_120
Yeah.	this is why we we started to look by having sort of voiced phonemes	Well , that 's the rea w w what I 'm arguing is that 's Yeah. I mean , uh , what I 'm arguing is that that that 's givi you gives you your intuition.	and Mm - hmm.	false	QMSum_120
this is why we we started to look by having sort of voiced phonemes	Well , that 's the rea w w what I 'm arguing is that 's Yeah. I mean , uh , what I 'm arguing is that that that 's givi you gives you your intuition.	and Mm - hmm.	But in in reality , it 's you know , there 's all of this this overlap and so forth ,	false	QMSum_120
Well , that 's the rea w w what I 'm arguing is that 's Yeah. I mean , uh , what I 'm arguing is that that that 's givi you gives you your intuition.	and Mm - hmm.	But in in reality , it 's you know , there 's all of this this overlap and so forth ,	Oh , sorry.	false	QMSum_120
and Mm - hmm.	But in in reality , it 's you know , there 's all of this this overlap and so forth ,	Oh , sorry.	and But what I 'm saying is that may be OK , because what you 're really getting is not actually voiced versus unvoiced , both for the fac the reason of the overlap and and then , uh , th you know , structural reasons , uh , uh , like the one that Chuck said , that that in fact , well , the data itself is that you 're working with is not perfect.	false	QMSum_120
But in in reality , it 's you know , there 's all of this this overlap and so forth ,	Oh , sorry.	and But what I 'm saying is that may be OK , because what you 're really getting is not actually voiced versus unvoiced , both for the fac the reason of the overlap and and then , uh , th you know , structural reasons , uh , uh , like the one that Chuck said , that that in fact , well , the data itself is that you 're working with is not perfect.	Yeah. Mm - hmm.	false	QMSum_120
Oh , sorry.	and But what I 'm saying is that may be OK , because what you 're really getting is not actually voiced versus unvoiced , both for the fac the reason of the overlap and and then , uh , th you know , structural reasons , uh , uh , like the one that Chuck said , that that in fact , well , the data itself is that you 're working with is not perfect.	Yeah. Mm - hmm.	So , what I 'm saying is maybe that 's not a killer because you 're just getting some characterization , one that 's driven by your intuition about voiced - unvoiced certainly ,	false	QMSum_120
and But what I 'm saying is that may be OK , because what you 're really getting is not actually voiced versus unvoiced , both for the fac the reason of the overlap and and then , uh , th you know , structural reasons , uh , uh , like the one that Chuck said , that that in fact , well , the data itself is that you 're working with is not perfect.	Yeah. Mm - hmm.	So , what I 'm saying is maybe that 's not a killer because you 're just getting some characterization , one that 's driven by your intuition about voiced - unvoiced certainly ,	Mm - hmm.	false	QMSum_120
Yeah. Mm - hmm.	So , what I 'm saying is maybe that 's not a killer because you 're just getting some characterization , one that 's driven by your intuition about voiced - unvoiced certainly ,	Mm - hmm.	but it 's just some characterization of something back in the in the in the almost raw data , rather than the smooth version.	false	QMSum_120
So , what I 'm saying is maybe that 's not a killer because you 're just getting some characterization , one that 's driven by your intuition about voiced - unvoiced certainly ,	Mm - hmm.	but it 's just some characterization of something back in the in the in the almost raw data , rather than the smooth version.	Mm - hmm.	false	QMSum_120
Mm - hmm.	but it 's just some characterization of something back in the in the in the almost raw data , rather than the smooth version.	Mm - hmm.	And your intuition is driving you towards particular kinds of , uh , statistical characterizations of , um , what 's missing from the spectral envelope.	false	QMSum_120
but it 's just some characterization of something back in the in the in the almost raw data , rather than the smooth version.	Mm - hmm.	And your intuition is driving you towards particular kinds of , uh , statistical characterizations of , um , what 's missing from the spectral envelope.	Mm - hmm.	false	QMSum_120
Mm - hmm.	And your intuition is driving you towards particular kinds of , uh , statistical characterizations of , um , what 's missing from the spectral envelope.	Mm - hmm.	Um , obviously you have something about the excitation , um , and what is it about the excitation , and , you know and you 're not getting the excitation anyway , you know. So so I I would almost take a uh , especially if if these trainings and so forth are faster , I would almost just take a uh , a scattershot at a few different ways of look of characterizing that difference and , uh , you could have one of them but and and see , you know , which of them helps.	false	QMSum_120
And your intuition is driving you towards particular kinds of , uh , statistical characterizations of , um , what 's missing from the spectral envelope.	Mm - hmm.	Um , obviously you have something about the excitation , um , and what is it about the excitation , and , you know and you 're not getting the excitation anyway , you know. So so I I would almost take a uh , especially if if these trainings and so forth are faster , I would almost just take a uh , a scattershot at a few different ways of look of characterizing that difference and , uh , you could have one of them but and and see , you know , which of them helps.	Mm - hmm. OK.	false	QMSum_120
Mm - hmm.	Um , obviously you have something about the excitation , um , and what is it about the excitation , and , you know and you 're not getting the excitation anyway , you know. So so I I would almost take a uh , especially if if these trainings and so forth are faster , I would almost just take a uh , a scattershot at a few different ways of look of characterizing that difference and , uh , you could have one of them but and and see , you know , which of them helps.	Mm - hmm. OK.	So i is the idea that you 're going to take whatever features you develop and and just add them onto the future vector ? Or , what 's the use of the the voiced - unvoiced detector ?	false	QMSum_120
Um , obviously you have something about the excitation , um , and what is it about the excitation , and , you know and you 're not getting the excitation anyway , you know. So so I I would almost take a uh , especially if if these trainings and so forth are faster , I would almost just take a uh , a scattershot at a few different ways of look of characterizing that difference and , uh , you could have one of them but and and see , you know , which of them helps.	Mm - hmm. OK.	So i is the idea that you 're going to take whatever features you develop and and just add them onto the future vector ? Or , what 's the use of the the voiced - unvoiced detector ?	Uh , I guess we don't know exactly yet. But , um Yeah. Th	false	QMSum_120
Mm - hmm. OK.	So i is the idea that you 're going to take whatever features you develop and and just add them onto the future vector ? Or , what 's the use of the the voiced - unvoiced detector ?	Uh , I guess we don't know exactly yet. But , um Yeah. Th	It 's not part of a VAD system that you 're doing ?	false	QMSum_120
So i is the idea that you 're going to take whatever features you develop and and just add them onto the future vector ? Or , what 's the use of the the voiced - unvoiced detector ?	Uh , I guess we don't know exactly yet. But , um Yeah. Th	It 's not part of a VAD system that you 're doing ?	No.	false	QMSum_120
Uh , I guess we don't know exactly yet. But , um Yeah. Th	It 's not part of a VAD system that you 're doing ?	No.	Uh , no. No.	false	QMSum_120
It 's not part of a VAD system that you 're doing ?	No.	Uh , no. No.	Oh , OK.	false	QMSum_120
No.	Uh , no. No.	Oh , OK.	No , the idea was , I guess , to to use them as as features.	false	QMSum_120
Uh , no. No.	Oh , OK.	No , the idea was , I guess , to to use them as as features.	Features. I see.	false	QMSum_120
Oh , OK.	No , the idea was , I guess , to to use them as as features.	Features. I see.	Uh Yeah , it could be , uh it could be a neural network that does voiced and unvoiced detection ,	false	QMSum_120
No , the idea was , I guess , to to use them as as features.	Features. I see.	Uh Yeah , it could be , uh it could be a neural network that does voiced and unvoiced detection ,	Mm - hmm.	false	QMSum_120
Features. I see.	Uh Yeah , it could be , uh it could be a neural network that does voiced and unvoiced detection ,	Mm - hmm.	but it could be in the also the big neural network that does phoneme classification.	false	QMSum_120
Uh Yeah , it could be , uh it could be a neural network that does voiced and unvoiced detection ,	Mm - hmm.	but it could be in the also the big neural network that does phoneme classification.	Mm - hmm.	false	QMSum_120
Mm - hmm.	but it could be in the also the big neural network that does phoneme classification.	Mm - hmm.	Mmm. Yeah.	false	QMSum_120
but it could be in the also the big neural network that does phoneme classification.	Mm - hmm.	Mmm. Yeah.	But each one of the mixture components I mean , you have , uh , uh , variance only , so it 's kind of like you 're just multiplying together these , um , probabilities from the individual features within each mixture. So it 's so , uh , it seems l you know	false	QMSum_120
Mm - hmm.	Mmm. Yeah.	But each one of the mixture components I mean , you have , uh , uh , variance only , so it 's kind of like you 're just multiplying together these , um , probabilities from the individual features within each mixture. So it 's so , uh , it seems l you know	I think it 's a neat thing. Uh , it seems like a good idea.	false	QMSum_120
Mmm. Yeah.	But each one of the mixture components I mean , you have , uh , uh , variance only , so it 's kind of like you 're just multiplying together these , um , probabilities from the individual features within each mixture. So it 's so , uh , it seems l you know	I think it 's a neat thing. Uh , it seems like a good idea.	Yeah. Um. Yeah. I mean , I know that , um , people doing some robustness things a ways back were were just doing just being gross and just throwing in the FFT and actually it wasn't wasn't wasn't so bad. Uh , so it would s and and you know that i it 's gotta hurt you a little bit to not have a a spectral , uh a s a smooth spectral envelope , so there must be something else that you get in return for that	false	QMSum_120
But each one of the mixture components I mean , you have , uh , uh , variance only , so it 's kind of like you 're just multiplying together these , um , probabilities from the individual features within each mixture. So it 's so , uh , it seems l you know	I think it 's a neat thing. Uh , it seems like a good idea.	Yeah. Um. Yeah. I mean , I know that , um , people doing some robustness things a ways back were were just doing just being gross and just throwing in the FFT and actually it wasn't wasn't wasn't so bad. Uh , so it would s and and you know that i it 's gotta hurt you a little bit to not have a a spectral , uh a s a smooth spectral envelope , so there must be something else that you get in return for that	Mm - hmm.	false	QMSum_120
I think it 's a neat thing. Uh , it seems like a good idea.	Yeah. Um. Yeah. I mean , I know that , um , people doing some robustness things a ways back were were just doing just being gross and just throwing in the FFT and actually it wasn't wasn't wasn't so bad. Uh , so it would s and and you know that i it 's gotta hurt you a little bit to not have a a spectral , uh a s a smooth spectral envelope , so there must be something else that you get in return for that	Mm - hmm.	that , uh uh So.	false	QMSum_120
Yeah. Um. Yeah. I mean , I know that , um , people doing some robustness things a ways back were were just doing just being gross and just throwing in the FFT and actually it wasn't wasn't wasn't so bad. Uh , so it would s and and you know that i it 's gotta hurt you a little bit to not have a a spectral , uh a s a smooth spectral envelope , so there must be something else that you get in return for that	Mm - hmm.	that , uh uh So.	So how does uh , maybe I 'm going in too much detail , but how exactly do you make the difference between the FFT and the smoothed spectral envelope ? Wha - wh i i uh , how is that , uh ?	false	QMSum_120
Mm - hmm.	that , uh uh So.	So how does uh , maybe I 'm going in too much detail , but how exactly do you make the difference between the FFT and the smoothed spectral envelope ? Wha - wh i i uh , how is that , uh ?	Um , we just How did we do it up again ?	false	QMSum_120
that , uh uh So.	So how does uh , maybe I 'm going in too much detail , but how exactly do you make the difference between the FFT and the smoothed spectral envelope ? Wha - wh i i uh , how is that , uh ?	Um , we just How did we do it up again ?	Uh , we distend the we have the twenty - three coefficient af after the mel f filter ,	false	QMSum_120
So how does uh , maybe I 'm going in too much detail , but how exactly do you make the difference between the FFT and the smoothed spectral envelope ? Wha - wh i i uh , how is that , uh ?	Um , we just How did we do it up again ?	Uh , we distend the we have the twenty - three coefficient af after the mel f filter ,	Mm - hmm.	false	QMSum_120

and

what sorts of features are you looking at ?

We have some

So we would be looking at , um , the variance of the spectrum of the excitation ,

false

QMSum_120

what sorts of features are you looking at ?

We have some

So we would be looking at , um , the variance of the spectrum of the excitation ,

uh , um , this , this , and this.

false

QMSum_120

We have some

So we would be looking at , um , the variance of the spectrum of the excitation ,

uh , um , this , this , and this.

something like this , which is should be high for voiced sounds. Uh , we

false

QMSum_120

So we would be looking at , um , the variance of the spectrum of the excitation ,

uh , um , this , this , and this.

something like this , which is should be high for voiced sounds. Uh , we

Wait a minute. I what does that mean ? The variance of the spectrum of excitation.

false

QMSum_120

uh , um , this , this , and this.

something like this , which is should be high for voiced sounds. Uh , we

Wait a minute. I what does that mean ? The variance of the spectrum of excitation.

Yeah. So the So basically the spectrum of the excitation for a purely periodic sig signal shou sh

false

QMSum_120

something like this , which is should be high for voiced sounds. Uh , we

Wait a minute. I what does that mean ? The variance of the spectrum of excitation.

Yeah. So the So basically the spectrum of the excitation for a purely periodic sig signal shou sh

OK. Yeah , w what yo what you 're calling the excitation , as I recall , is you 're subtracting the the , um the mel mel mel filter , uh , spectrum from the FFT spectrum.

false

QMSum_120

Wait a minute. I what does that mean ? The variance of the spectrum of excitation.

Yeah. So the So basically the spectrum of the excitation for a purely periodic sig signal shou sh

OK. Yeah , w what yo what you 're calling the excitation , as I recall , is you 're subtracting the the , um the mel mel mel filter , uh , spectrum from the FFT spectrum.

e That 's right. Yeah. So

false

QMSum_120

Yeah. So the So basically the spectrum of the excitation for a purely periodic sig signal shou sh

OK. Yeah , w what yo what you 're calling the excitation , as I recall , is you 're subtracting the the , um the mel mel mel filter , uh , spectrum from the FFT spectrum.

e That 's right. Yeah. So

Right.

false

QMSum_120

OK. Yeah , w what yo what you 're calling the excitation , as I recall , is you 're subtracting the the , um the mel mel mel filter , uh , spectrum from the FFT spectrum.

e That 's right. Yeah. So

Right.

Yeah.

false

QMSum_120

e That 's right. Yeah. So

Right.

Yeah.

Mm - hmm.

false

QMSum_120

Right.

Yeah.

Mm - hmm.

So we have the mel f filter bank , we have the FFT , so we just

false

QMSum_120

Yeah.

Mm - hmm.

So we have the mel f filter bank , we have the FFT , so we just

So it 's it 's not really an excitation ,

false

QMSum_120

Mm - hmm.

So we have the mel f filter bank , we have the FFT , so we just

So it 's it 's not really an excitation ,

No.

false

QMSum_120

So we have the mel f filter bank , we have the FFT , so we just

So it 's it 's not really an excitation ,

No.

but it 's something that hopefully tells you something about the excitation.

false

QMSum_120

So it 's it 's not really an excitation ,

No.

but it 's something that hopefully tells you something about the excitation.

Yeah , that 's right.

false

QMSum_120

No.

but it 's something that hopefully tells you something about the excitation.

Yeah , that 's right.

Yeah , yeah.

false

QMSum_120

but it 's something that hopefully tells you something about the excitation.

Yeah , that 's right.

Yeah , yeah.

Um Yeah.

false

QMSum_120

Yeah , that 's right.

Yeah , yeah.

Um Yeah.

We have here some histogram ,

false

QMSum_120

Yeah , yeah.

Um Yeah.

We have here some histogram ,

E yeah ,

false

QMSum_120

Um Yeah.

We have here some histogram ,

E yeah ,

but they have a lot of overlap.

false

QMSum_120

We have here some histogram ,

E yeah ,

but they have a lot of overlap.

but it 's it 's still Yeah. So , well , for unvoiced portion we have something tha that has a mean around O point three , and for voiced portion the mean is O point fifty - nine. But the variance seem quite high.

false

QMSum_120

E yeah ,

but they have a lot of overlap.

but it 's it 's still Yeah. So , well , for unvoiced portion we have something tha that has a mean around O point three , and for voiced portion the mean is O point fifty - nine. But the variance seem quite high.

How do you know ?

false

QMSum_120

but they have a lot of overlap.

but it 's it 's still Yeah. So , well , for unvoiced portion we have something tha that has a mean around O point three , and for voiced portion the mean is O point fifty - nine. But the variance seem quite high.

How do you know ?

So Mmm.

false

QMSum_120

but it 's it 's still Yeah. So , well , for unvoiced portion we have something tha that has a mean around O point three , and for voiced portion the mean is O point fifty - nine. But the variance seem quite high.

How do you know ?

So Mmm.

How did you get your voiced and unvoiced truth data ?

false

QMSum_120

How do you know ?

So Mmm.

How did you get your voiced and unvoiced truth data ?

We used , uh , TIMIT and we used canonical mappings between the phones

false

QMSum_120

So Mmm.

How did you get your voiced and unvoiced truth data ?

We used , uh , TIMIT and we used canonical mappings between the phones

Yeah. We , uh , use TIMIT on this ,

false

QMSum_120

How did you get your voiced and unvoiced truth data ?

We used , uh , TIMIT and we used canonical mappings between the phones

Yeah. We , uh , use TIMIT on this ,

and

false

QMSum_120

We used , uh , TIMIT and we used canonical mappings between the phones

Yeah. We , uh , use TIMIT on this ,

and

for

false

QMSum_120

Yeah. We , uh , use TIMIT on this ,

and

for

th Yeah.

false

QMSum_120

and

for

th Yeah.

But if we look at it in one sentence , it apparently it 's good , I think.

false

QMSum_120

for

th Yeah.

But if we look at it in one sentence , it apparently it 's good , I think.

Yeah , but Yeah. Uh , so it 's noisy TIMIT. That 's right. Yeah.

false

QMSum_120

th Yeah.

But if we look at it in one sentence , it apparently it 's good , I think.

Yeah , but Yeah. Uh , so it 's noisy TIMIT. That 's right. Yeah.

It 's noisy TIMIT.

false

QMSum_120

But if we look at it in one sentence , it apparently it 's good , I think.

Yeah , but Yeah. Uh , so it 's noisy TIMIT. That 's right. Yeah.

It 's noisy TIMIT.

Yeah.

false

QMSum_120

Yeah , but Yeah. Uh , so it 's noisy TIMIT. That 's right. Yeah.

It 's noisy TIMIT.

Yeah.

It seems quite robust to noise , so when we take we draw its parameters across time for a clean sentence and then nois the same noisy sentence , it 's very close.

false

QMSum_120

It 's noisy TIMIT.

Yeah.

It seems quite robust to noise , so when we take we draw its parameters across time for a clean sentence and then nois the same noisy sentence , it 's very close.

Mm - hmm.

false

QMSum_120

Yeah.

It seems quite robust to noise , so when we take we draw its parameters across time for a clean sentence and then nois the same noisy sentence , it 's very close.

Mm - hmm.

Yeah. So there are there is this. There could be also the , um something like the maximum of the auto - correlation function or which

false

QMSum_120

It seems quite robust to noise , so when we take we draw its parameters across time for a clean sentence and then nois the same noisy sentence , it 's very close.

Mm - hmm.

Yeah. So there are there is this. There could be also the , um something like the maximum of the auto - correlation function or which

Is this a a s a trained system ? Or is it a system where you just pick some thresholds ? Ho - how does it work ?

false

QMSum_120

Mm - hmm.

Yeah. So there are there is this. There could be also the , um something like the maximum of the auto - correlation function or which

Is this a a s a trained system ? Or is it a system where you just pick some thresholds ? Ho - how does it work ?

Right now we just are trying to find some features. And ,

false

QMSum_120

Yeah. So there are there is this. There could be also the , um something like the maximum of the auto - correlation function or which

Is this a a s a trained system ? Or is it a system where you just pick some thresholds ? Ho - how does it work ?

Right now we just are trying to find some features. And ,

Mm - hmm.

false

QMSum_120

Is this a a s a trained system ? Or is it a system where you just pick some thresholds ? Ho - how does it work ?

Right now we just are trying to find some features. And ,

Mm - hmm.

uh Yeah. Hopefully , I think what we want to have is to put these features in s some kind of , um well , to to obtain a statistical model on these features and to or just to use a neural network and hopefully these features w would help

false

QMSum_120

Right now we just are trying to find some features. And ,

Mm - hmm.

uh Yeah. Hopefully , I think what we want to have is to put these features in s some kind of , um well , to to obtain a statistical model on these features and to or just to use a neural network and hopefully these features w would help

Because it seems like what you said about the mean of the the voiced and the unvoiced that seemed pretty encouraging.

false

QMSum_120

Mm - hmm.

uh Yeah. Hopefully , I think what we want to have is to put these features in s some kind of , um well , to to obtain a statistical model on these features and to or just to use a neural network and hopefully these features w would help

Because it seems like what you said about the mean of the the voiced and the unvoiced that seemed pretty encouraging.

Mm - hmm.

false

QMSum_120

uh Yeah. Hopefully , I think what we want to have is to put these features in s some kind of , um well , to to obtain a statistical model on these features and to or just to use a neural network and hopefully these features w would help

Because it seems like what you said about the mean of the the voiced and the unvoiced that seemed pretty encouraging.

Mm - hmm.

Well , yeah , except the variance was big.

false

QMSum_120

Because it seems like what you said about the mean of the the voiced and the unvoiced that seemed pretty encouraging.

Mm - hmm.

Well , yeah , except the variance was big.

Right ?

false

QMSum_120

Mm - hmm.

Well , yeah , except the variance was big.

Right ?

Yeah. Except the variance is quite high.

false

QMSum_120

Well , yeah , except the variance was big.

Right ?

Yeah. Except the variance is quite high.

Right ?

false

QMSum_120

Right ?

Yeah. Except the variance is quite high.

Right ?

Well , y

false

QMSum_120

Yeah. Except the variance is quite high.

Right ?

Well , y

Yeah.

false

QMSum_120

Right ?

Well , y

Yeah.

Well , y I I don't know that I would trust that so much because you 're doing these canonical mappings from TIMIT labellings.

false

QMSum_120

Well , y

Yeah.

Well , y I I don't know that I would trust that so much because you 're doing these canonical mappings from TIMIT labellings.

Uh - huh.

false

QMSum_120

Yeah.

Well , y I I don't know that I would trust that so much because you 're doing these canonical mappings from TIMIT labellings.

Uh - huh.

Right ? So , really that 's sort of a cartoon picture about what 's voiced and unvoiced. So that could be giving you a lot of variance.

false

QMSum_120

Well , y I I don't know that I would trust that so much because you 're doing these canonical mappings from TIMIT labellings.

Uh - huh.

Right ? So , really that 's sort of a cartoon picture about what 's voiced and unvoiced. So that could be giving you a lot of variance.

Yeah.

false

QMSum_120

Uh - huh.

Right ? So , really that 's sort of a cartoon picture about what 's voiced and unvoiced. So that could be giving you a lot of variance.

Yeah.

I mean , i it it may be that that you 're finding something good and that the variance is sort of artificial because of how you 're getting your truth.

false

QMSum_120

Right ? So , really that 's sort of a cartoon picture about what 's voiced and unvoiced. So that could be giving you a lot of variance.

Yeah.

I mean , i it it may be that that you 're finding something good and that the variance is sort of artificial because of how you 're getting your truth.

Mm - hmm.

false

QMSum_120

Yeah.

I mean , i it it may be that that you 're finding something good and that the variance is sort of artificial because of how you 're getting your truth.

Mm - hmm.

Yeah. But another way of looking at it might be that I mean , what w we we are coming up with feature sets after all. So another way of looking at it is that um , the mel cepstru mel spectrum , mel cepstrum , any of these variants , um , give you the smooth spectrum. It 's the spectral envelope. By going back to the FFT , you 're getting something that is more like the raw data. So the question is , what characterization and you 're playing around with this another way of looking at it is what characterization of the difference between the raw data and this smooth version is something that you 're missing that could help ? So , I mean , looking at different statistical measures of that difference , coming up with some things and just trying them out and seeing if you add them onto the feature vector does that make things better or worse in noise , where you 're really just i i the way I 'm looking at it is not so much you 're trying to f find the best the world 's best voiced - unvoiced , uh , uh , classifier ,

false

QMSum_120

I mean , i it it may be that that you 're finding something good and that the variance is sort of artificial because of how you 're getting your truth.

Mm - hmm.

Yeah. But another way of looking at it might be that I mean , what w we we are coming up with feature sets after all. So another way of looking at it is that um , the mel cepstru mel spectrum , mel cepstrum , any of these variants , um , give you the smooth spectrum. It 's the spectral envelope. By going back to the FFT , you 're getting something that is more like the raw data. So the question is , what characterization and you 're playing around with this another way of looking at it is what characterization of the difference between the raw data and this smooth version is something that you 're missing that could help ? So , I mean , looking at different statistical measures of that difference , coming up with some things and just trying them out and seeing if you add them onto the feature vector does that make things better or worse in noise , where you 're really just i i the way I 'm looking at it is not so much you 're trying to f find the best the world 's best voiced - unvoiced , uh , uh , classifier ,

Mm - hmm.

false

QMSum_120

Mm - hmm.

Yeah. But another way of looking at it might be that I mean , what w we we are coming up with feature sets after all. So another way of looking at it is that um , the mel cepstru mel spectrum , mel cepstrum , any of these variants , um , give you the smooth spectrum. It 's the spectral envelope. By going back to the FFT , you 're getting something that is more like the raw data. So the question is , what characterization and you 're playing around with this another way of looking at it is what characterization of the difference between the raw data and this smooth version is something that you 're missing that could help ? So , I mean , looking at different statistical measures of that difference , coming up with some things and just trying them out and seeing if you add them onto the feature vector does that make things better or worse in noise , where you 're really just i i the way I 'm looking at it is not so much you 're trying to f find the best the world 's best voiced - unvoiced , uh , uh , classifier ,

Mm - hmm.

Mmm.

false

QMSum_120

Yeah. But another way of looking at it might be that I mean , what w we we are coming up with feature sets after all. So another way of looking at it is that um , the mel cepstru mel spectrum , mel cepstrum , any of these variants , um , give you the smooth spectrum. It 's the spectral envelope. By going back to the FFT , you 're getting something that is more like the raw data. So the question is , what characterization and you 're playing around with this another way of looking at it is what characterization of the difference between the raw data and this smooth version is something that you 're missing that could help ? So , I mean , looking at different statistical measures of that difference , coming up with some things and just trying them out and seeing if you add them onto the feature vector does that make things better or worse in noise , where you 're really just i i the way I 'm looking at it is not so much you 're trying to f find the best the world 's best voiced - unvoiced , uh , uh , classifier ,

Mm - hmm.

Mmm.

but it 's more that , you know , uh , uh , try some different statistical characterizations of that difference back to the raw data

false

QMSum_120

Mm - hmm.

Mmm.

but it 's more that , you know , uh , uh , try some different statistical characterizations of that difference back to the raw data

Right.

false

QMSum_120

Mmm.

but it 's more that , you know , uh , uh , try some different statistical characterizations of that difference back to the raw data

Right.

and and m maybe there 's something there that the system can use.

false

QMSum_120

but it 's more that , you know , uh , uh , try some different statistical characterizations of that difference back to the raw data

Right.

and and m maybe there 's something there that the system can use.

Right.

false

QMSum_120

Right.

and and m maybe there 's something there that the system can use.

Right.

Yeah. Yeah , but ther more obvious is that Yeah. The the more obvious is that that well , using the th the FFT , um , you just it gives you just information about if it 's voiced or not voiced , ma mainly , I mean. But So ,

false

QMSum_120

and and m maybe there 's something there that the system can use.

Right.

Yeah. Yeah , but ther more obvious is that Yeah. The the more obvious is that that well , using the th the FFT , um , you just it gives you just information about if it 's voiced or not voiced , ma mainly , I mean. But So ,

Yeah.

false

QMSum_120

Right.

Yeah. Yeah , but ther more obvious is that Yeah. The the more obvious is that that well , using the th the FFT , um , you just it gives you just information about if it 's voiced or not voiced , ma mainly , I mean. But So ,

Yeah.

this is why we we started to look by having sort of voiced phonemes

false

QMSum_120

Yeah. Yeah , but ther more obvious is that Yeah. The the more obvious is that that well , using the th the FFT , um , you just it gives you just information about if it 's voiced or not voiced , ma mainly , I mean. But So ,

Yeah.

this is why we we started to look by having sort of voiced phonemes

Well , that 's the rea w w what I 'm arguing is that 's Yeah. I mean , uh , what I 'm arguing is that that that 's givi you gives you your intuition.

false

QMSum_120

Yeah.

this is why we we started to look by having sort of voiced phonemes

Well , that 's the rea w w what I 'm arguing is that 's Yeah. I mean , uh , what I 'm arguing is that that that 's givi you gives you your intuition.

and Mm - hmm.

false

QMSum_120

this is why we we started to look by having sort of voiced phonemes

Well , that 's the rea w w what I 'm arguing is that 's Yeah. I mean , uh , what I 'm arguing is that that that 's givi you gives you your intuition.

and Mm - hmm.

But in in reality , it 's you know , there 's all of this this overlap and so forth ,

false

QMSum_120

Well , that 's the rea w w what I 'm arguing is that 's Yeah. I mean , uh , what I 'm arguing is that that that 's givi you gives you your intuition.

and Mm - hmm.

But in in reality , it 's you know , there 's all of this this overlap and so forth ,

Oh , sorry.

false

QMSum_120

and Mm - hmm.

But in in reality , it 's you know , there 's all of this this overlap and so forth ,

Oh , sorry.

and But what I 'm saying is that may be OK , because what you 're really getting is not actually voiced versus unvoiced , both for the fac the reason of the overlap and and then , uh , th you know , structural reasons , uh , uh , like the one that Chuck said , that that in fact , well , the data itself is that you 're working with is not perfect.

false

QMSum_120

But in in reality , it 's you know , there 's all of this this overlap and so forth ,

Oh , sorry.

and But what I 'm saying is that may be OK , because what you 're really getting is not actually voiced versus unvoiced , both for the fac the reason of the overlap and and then , uh , th you know , structural reasons , uh , uh , like the one that Chuck said , that that in fact , well , the data itself is that you 're working with is not perfect.

Yeah. Mm - hmm.

false

QMSum_120

Oh , sorry.

and But what I 'm saying is that may be OK , because what you 're really getting is not actually voiced versus unvoiced , both for the fac the reason of the overlap and and then , uh , th you know , structural reasons , uh , uh , like the one that Chuck said , that that in fact , well , the data itself is that you 're working with is not perfect.

Yeah. Mm - hmm.

So , what I 'm saying is maybe that 's not a killer because you 're just getting some characterization , one that 's driven by your intuition about voiced - unvoiced certainly ,

false

QMSum_120

and But what I 'm saying is that may be OK , because what you 're really getting is not actually voiced versus unvoiced , both for the fac the reason of the overlap and and then , uh , th you know , structural reasons , uh , uh , like the one that Chuck said , that that in fact , well , the data itself is that you 're working with is not perfect.

Yeah. Mm - hmm.

So , what I 'm saying is maybe that 's not a killer because you 're just getting some characterization , one that 's driven by your intuition about voiced - unvoiced certainly ,

Mm - hmm.

false

QMSum_120

Yeah. Mm - hmm.

So , what I 'm saying is maybe that 's not a killer because you 're just getting some characterization , one that 's driven by your intuition about voiced - unvoiced certainly ,

Mm - hmm.

but it 's just some characterization of something back in the in the in the almost raw data , rather than the smooth version.

false

QMSum_120

So , what I 'm saying is maybe that 's not a killer because you 're just getting some characterization , one that 's driven by your intuition about voiced - unvoiced certainly ,

Mm - hmm.

but it 's just some characterization of something back in the in the in the almost raw data , rather than the smooth version.

Mm - hmm.

false

QMSum_120

Mm - hmm.

but it 's just some characterization of something back in the in the in the almost raw data , rather than the smooth version.

Mm - hmm.

And your intuition is driving you towards particular kinds of , uh , statistical characterizations of , um , what 's missing from the spectral envelope.

false

QMSum_120

but it 's just some characterization of something back in the in the in the almost raw data , rather than the smooth version.

Mm - hmm.

And your intuition is driving you towards particular kinds of , uh , statistical characterizations of , um , what 's missing from the spectral envelope.

Mm - hmm.

false

QMSum_120

Mm - hmm.

And your intuition is driving you towards particular kinds of , uh , statistical characterizations of , um , what 's missing from the spectral envelope.

Mm - hmm.

Um , obviously you have something about the excitation , um , and what is it about the excitation , and , you know and you 're not getting the excitation anyway , you know. So so I I would almost take a uh , especially if if these trainings and so forth are faster , I would almost just take a uh , a scattershot at a few different ways of look of characterizing that difference and , uh , you could have one of them but and and see , you know , which of them helps.

false

QMSum_120

And your intuition is driving you towards particular kinds of , uh , statistical characterizations of , um , what 's missing from the spectral envelope.

Mm - hmm.

Um , obviously you have something about the excitation , um , and what is it about the excitation , and , you know and you 're not getting the excitation anyway , you know. So so I I would almost take a uh , especially if if these trainings and so forth are faster , I would almost just take a uh , a scattershot at a few different ways of look of characterizing that difference and , uh , you could have one of them but and and see , you know , which of them helps.

Mm - hmm. OK.

false

QMSum_120

Mm - hmm.

Um , obviously you have something about the excitation , um , and what is it about the excitation , and , you know and you 're not getting the excitation anyway , you know. So so I I would almost take a uh , especially if if these trainings and so forth are faster , I would almost just take a uh , a scattershot at a few different ways of look of characterizing that difference and , uh , you could have one of them but and and see , you know , which of them helps.

Mm - hmm. OK.

So i is the idea that you 're going to take whatever features you develop and and just add them onto the future vector ? Or , what 's the use of the the voiced - unvoiced detector ?

false

QMSum_120

Um , obviously you have something about the excitation , um , and what is it about the excitation , and , you know and you 're not getting the excitation anyway , you know. So so I I would almost take a uh , especially if if these trainings and so forth are faster , I would almost just take a uh , a scattershot at a few different ways of look of characterizing that difference and , uh , you could have one of them but and and see , you know , which of them helps.

Mm - hmm. OK.

So i is the idea that you 're going to take whatever features you develop and and just add them onto the future vector ? Or , what 's the use of the the voiced - unvoiced detector ?

Uh , I guess we don't know exactly yet. But , um Yeah. Th

false

QMSum_120

Mm - hmm. OK.

So i is the idea that you 're going to take whatever features you develop and and just add them onto the future vector ? Or , what 's the use of the the voiced - unvoiced detector ?

Uh , I guess we don't know exactly yet. But , um Yeah. Th

It 's not part of a VAD system that you 're doing ?

false

QMSum_120

So i is the idea that you 're going to take whatever features you develop and and just add them onto the future vector ? Or , what 's the use of the the voiced - unvoiced detector ?

Uh , I guess we don't know exactly yet. But , um Yeah. Th

It 's not part of a VAD system that you 're doing ?

No.

false

QMSum_120

Uh , I guess we don't know exactly yet. But , um Yeah. Th

It 's not part of a VAD system that you 're doing ?

No.

Uh , no. No.

false

QMSum_120

It 's not part of a VAD system that you 're doing ?

No.

Uh , no. No.

Oh , OK.

false

QMSum_120

No.

Uh , no. No.

Oh , OK.

No , the idea was , I guess , to to use them as as features.

false

QMSum_120

Uh , no. No.

Oh , OK.

No , the idea was , I guess , to to use them as as features.

Features. I see.

false

QMSum_120

Oh , OK.

No , the idea was , I guess , to to use them as as features.

Features. I see.

Uh Yeah , it could be , uh it could be a neural network that does voiced and unvoiced detection ,

false

QMSum_120

No , the idea was , I guess , to to use them as as features.

Features. I see.

Uh Yeah , it could be , uh it could be a neural network that does voiced and unvoiced detection ,

Mm - hmm.

false

QMSum_120

Features. I see.

Uh Yeah , it could be , uh it could be a neural network that does voiced and unvoiced detection ,

Mm - hmm.

but it could be in the also the big neural network that does phoneme classification.

false

QMSum_120

Uh Yeah , it could be , uh it could be a neural network that does voiced and unvoiced detection ,

Mm - hmm.

but it could be in the also the big neural network that does phoneme classification.

Mm - hmm.

false

QMSum_120

Mm - hmm.

but it could be in the also the big neural network that does phoneme classification.

Mm - hmm.

Mmm. Yeah.

false

QMSum_120

but it could be in the also the big neural network that does phoneme classification.

Mm - hmm.

Mmm. Yeah.

But each one of the mixture components I mean , you have , uh , uh , variance only , so it 's kind of like you 're just multiplying together these , um , probabilities from the individual features within each mixture. So it 's so , uh , it seems l you know

false

QMSum_120

Mm - hmm.

Mmm. Yeah.

But each one of the mixture components I mean , you have , uh , uh , variance only , so it 's kind of like you 're just multiplying together these , um , probabilities from the individual features within each mixture. So it 's so , uh , it seems l you know

I think it 's a neat thing. Uh , it seems like a good idea.

false

QMSum_120

Mmm. Yeah.

But each one of the mixture components I mean , you have , uh , uh , variance only , so it 's kind of like you 're just multiplying together these , um , probabilities from the individual features within each mixture. So it 's so , uh , it seems l you know

I think it 's a neat thing. Uh , it seems like a good idea.

Yeah. Um. Yeah. I mean , I know that , um , people doing some robustness things a ways back were were just doing just being gross and just throwing in the FFT and actually it wasn't wasn't wasn't so bad. Uh , so it would s and and you know that i it 's gotta hurt you a little bit to not have a a spectral , uh a s a smooth spectral envelope , so there must be something else that you get in return for that

false

QMSum_120

But each one of the mixture components I mean , you have , uh , uh , variance only , so it 's kind of like you 're just multiplying together these , um , probabilities from the individual features within each mixture. So it 's so , uh , it seems l you know

I think it 's a neat thing. Uh , it seems like a good idea.

Yeah. Um. Yeah. I mean , I know that , um , people doing some robustness things a ways back were were just doing just being gross and just throwing in the FFT and actually it wasn't wasn't wasn't so bad. Uh , so it would s and and you know that i it 's gotta hurt you a little bit to not have a a spectral , uh a s a smooth spectral envelope , so there must be something else that you get in return for that

Mm - hmm.

false

QMSum_120

I think it 's a neat thing. Uh , it seems like a good idea.

Yeah. Um. Yeah. I mean , I know that , um , people doing some robustness things a ways back were were just doing just being gross and just throwing in the FFT and actually it wasn't wasn't wasn't so bad. Uh , so it would s and and you know that i it 's gotta hurt you a little bit to not have a a spectral , uh a s a smooth spectral envelope , so there must be something else that you get in return for that

Mm - hmm.

that , uh uh So.

false

QMSum_120

Yeah. Um. Yeah. I mean , I know that , um , people doing some robustness things a ways back were were just doing just being gross and just throwing in the FFT and actually it wasn't wasn't wasn't so bad. Uh , so it would s and and you know that i it 's gotta hurt you a little bit to not have a a spectral , uh a s a smooth spectral envelope , so there must be something else that you get in return for that

Mm - hmm.

that , uh uh So.

So how does uh , maybe I 'm going in too much detail , but how exactly do you make the difference between the FFT and the smoothed spectral envelope ? Wha - wh i i uh , how is that , uh ?

false

QMSum_120

Mm - hmm.

that , uh uh So.

So how does uh , maybe I 'm going in too much detail , but how exactly do you make the difference between the FFT and the smoothed spectral envelope ? Wha - wh i i uh , how is that , uh ?

Um , we just How did we do it up again ?

false

QMSum_120

that , uh uh So.

So how does uh , maybe I 'm going in too much detail , but how exactly do you make the difference between the FFT and the smoothed spectral envelope ? Wha - wh i i uh , how is that , uh ?

Um , we just How did we do it up again ?

Uh , we distend the we have the twenty - three coefficient af after the mel f filter ,

false

QMSum_120

So how does uh , maybe I 'm going in too much detail , but how exactly do you make the difference between the FFT and the smoothed spectral envelope ? Wha - wh i i uh , how is that , uh ?

Um , we just How did we do it up again ?

Uh , we distend the we have the twenty - three coefficient af after the mel f filter ,

Mm - hmm.

false

QMSum_120