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
|
|---|---|---|---|---|---|
Yeah , it compiled fine actually.
|
Yeah.
|
No no errors. Nothing. So.
|
Uh , this is slightly off topic
| false |
QMSum_194
|
Yeah.
|
No no errors. Nothing. So.
|
Uh , this is slightly off topic
|
That 's good.
| false |
QMSum_194
|
No no errors. Nothing. So.
|
Uh , this is slightly off topic
|
That 's good.
|
but , uh , I noticed , just glancing at the , uh , Hopkins workshop , uh , web site that , uh , um one of the thing I don't know Well , we 'll see how much they accomplish , but one of the things that they were trying to do in the graphical models thing was to put together a a , uh , tool kit for doing , uh r um , arbitrary graphical models for , uh , speech recognition.
| false |
QMSum_194
|
Uh , this is slightly off topic
|
That 's good.
|
but , uh , I noticed , just glancing at the , uh , Hopkins workshop , uh , web site that , uh , um one of the thing I don't know Well , we 'll see how much they accomplish , but one of the things that they were trying to do in the graphical models thing was to put together a a , uh , tool kit for doing , uh r um , arbitrary graphical models for , uh , speech recognition.
|
Hmm.
| false |
QMSum_194
|
That 's good.
|
but , uh , I noticed , just glancing at the , uh , Hopkins workshop , uh , web site that , uh , um one of the thing I don't know Well , we 'll see how much they accomplish , but one of the things that they were trying to do in the graphical models thing was to put together a a , uh , tool kit for doing , uh r um , arbitrary graphical models for , uh , speech recognition.
|
Hmm.
|
So And Jeff , uh the two Jeffs were
| false |
QMSum_194
|
but , uh , I noticed , just glancing at the , uh , Hopkins workshop , uh , web site that , uh , um one of the thing I don't know Well , we 'll see how much they accomplish , but one of the things that they were trying to do in the graphical models thing was to put together a a , uh , tool kit for doing , uh r um , arbitrary graphical models for , uh , speech recognition.
|
Hmm.
|
So And Jeff , uh the two Jeffs were
|
Who 's the second Jeff ?
| false |
QMSum_194
|
Hmm.
|
So And Jeff , uh the two Jeffs were
|
Who 's the second Jeff ?
|
Uh Oh , uh , do you know Geoff Zweig ?
| false |
QMSum_194
|
So And Jeff , uh the two Jeffs were
|
Who 's the second Jeff ?
|
Uh Oh , uh , do you know Geoff Zweig ?
|
No.
| false |
QMSum_194
|
Who 's the second Jeff ?
|
Uh Oh , uh , do you know Geoff Zweig ?
|
No.
|
Oh. Uh , he he , uh he was here for a couple years
| false |
QMSum_194
|
Uh Oh , uh , do you know Geoff Zweig ?
|
No.
|
Oh. Uh , he he , uh he was here for a couple years
|
Oh , OK.
| false |
QMSum_194
|
No.
|
Oh. Uh , he he , uh he was here for a couple years
|
Oh , OK.
|
and he , uh got his PHD. He And he 's , uh , been at IBM for the last couple years.
| false |
QMSum_194
|
Oh. Uh , he he , uh he was here for a couple years
|
Oh , OK.
|
and he , uh got his PHD. He And he 's , uh , been at IBM for the last couple years.
|
Oh , OK.
| false |
QMSum_194
|
Oh , OK.
|
and he , uh got his PHD. He And he 's , uh , been at IBM for the last couple years.
|
Oh , OK.
|
So.
| false |
QMSum_194
|
and he , uh got his PHD. He And he 's , uh , been at IBM for the last couple years.
|
Oh , OK.
|
So.
|
Wow. That would be neat.
| false |
QMSum_194
|
Oh , OK.
|
So.
|
Wow. That would be neat.
|
Uh , so he did he did his PHD on dynamic Bayes - nets , uh , for for speech recognition. He had some continuity built into the model , presumably to handle some , um , inertia in the in the production system , and , um
| false |
QMSum_194
|
So.
|
Wow. That would be neat.
|
Uh , so he did he did his PHD on dynamic Bayes - nets , uh , for for speech recognition. He had some continuity built into the model , presumably to handle some , um , inertia in the in the production system , and , um
|
Hmm.
| false |
QMSum_194
|
Wow. That would be neat.
|
Uh , so he did he did his PHD on dynamic Bayes - nets , uh , for for speech recognition. He had some continuity built into the model , presumably to handle some , um , inertia in the in the production system , and , um
|
Hmm.
|
So.
| false |
QMSum_194
|
Uh , so he did he did his PHD on dynamic Bayes - nets , uh , for for speech recognition. He had some continuity built into the model , presumably to handle some , um , inertia in the in the production system , and , um
|
Hmm.
|
So.
|
Hmm.
| false |
QMSum_194
|
Hmm.
|
So.
|
Hmm.
|
Um , I 've been playing with , first , the , um , VAD. Um , so it 's exactly the same approach , but the features that the VAD neural network use are , uh , MFCC after noise compensation. Oh , I think I have the results.
| false |
QMSum_194
|
So.
|
Hmm.
|
Um , I 've been playing with , first , the , um , VAD. Um , so it 's exactly the same approach , but the features that the VAD neural network use are , uh , MFCC after noise compensation. Oh , I think I have the results.
|
What was it using before ?
| true |
QMSum_194
|
Hmm.
|
Um , I 've been playing with , first , the , um , VAD. Um , so it 's exactly the same approach , but the features that the VAD neural network use are , uh , MFCC after noise compensation. Oh , I think I have the results.
|
What was it using before ?
|
Before it was just P L
| false |
QMSum_194
|
Um , I 've been playing with , first , the , um , VAD. Um , so it 's exactly the same approach , but the features that the VAD neural network use are , uh , MFCC after noise compensation. Oh , I think I have the results.
|
What was it using before ?
|
Before it was just P L
|
So.
| false |
QMSum_194
|
What was it using before ?
|
Before it was just P L
|
So.
|
Yeah , it was actually No. Not I mean , it was just the noisy features I guess.
| false |
QMSum_194
|
Before it was just P L
|
So.
|
Yeah , it was actually No. Not I mean , it was just the noisy features I guess.
|
Yeah ,
| false |
QMSum_194
|
So.
|
Yeah , it was actually No. Not I mean , it was just the noisy features I guess.
|
Yeah ,
|
Yeah , yeah , yeah ,
| false |
QMSum_194
|
Yeah , it was actually No. Not I mean , it was just the noisy features I guess.
|
Yeah ,
|
Yeah , yeah , yeah ,
|
noisy noisy features.
| false |
QMSum_194
|
Yeah ,
|
Yeah , yeah , yeah ,
|
noisy noisy features.
|
not compensated.
| false |
QMSum_194
|
Yeah , yeah , yeah ,
|
noisy noisy features.
|
not compensated.
|
Um This is what we get after This So , actually , we , yeah , here the features are noise compensated and there is also the LDA filter. Um , and then it 's a pretty small neural network which use , um , nine frames of of six features from C - zero to C - fives , plus the first derivatives. And it has one hundred hidden units.
| false |
QMSum_194
|
noisy noisy features.
|
not compensated.
|
Um This is what we get after This So , actually , we , yeah , here the features are noise compensated and there is also the LDA filter. Um , and then it 's a pretty small neural network which use , um , nine frames of of six features from C - zero to C - fives , plus the first derivatives. And it has one hundred hidden units.
|
Is that nine frames u s uh , centered around the current frame ? Or
| false |
QMSum_194
|
not compensated.
|
Um This is what we get after This So , actually , we , yeah , here the features are noise compensated and there is also the LDA filter. Um , and then it 's a pretty small neural network which use , um , nine frames of of six features from C - zero to C - fives , plus the first derivatives. And it has one hundred hidden units.
|
Is that nine frames u s uh , centered around the current frame ? Or
|
Yeah. Mm - hmm.
| false |
QMSum_194
|
Um This is what we get after This So , actually , we , yeah , here the features are noise compensated and there is also the LDA filter. Um , and then it 's a pretty small neural network which use , um , nine frames of of six features from C - zero to C - fives , plus the first derivatives. And it has one hundred hidden units.
|
Is that nine frames u s uh , centered around the current frame ? Or
|
Yeah. Mm - hmm.
|
S so , I 'm I 'm sorry , there 's there 's there 's how many how many inputs ?
| false |
QMSum_194
|
Is that nine frames u s uh , centered around the current frame ? Or
|
Yeah. Mm - hmm.
|
S so , I 'm I 'm sorry , there 's there 's there 's how many how many inputs ?
|
So it 's twelve times nine.
| false |
QMSum_194
|
Yeah. Mm - hmm.
|
S so , I 'm I 'm sorry , there 's there 's there 's how many how many inputs ?
|
So it 's twelve times nine.
|
Twelve times nine inputs , and a hundred , uh , hidden.
| false |
QMSum_194
|
S so , I 'm I 'm sorry , there 's there 's there 's how many how many inputs ?
|
So it 's twelve times nine.
|
Twelve times nine inputs , and a hundred , uh , hidden.
|
Hidden and
| false |
QMSum_194
|
So it 's twelve times nine.
|
Twelve times nine inputs , and a hundred , uh , hidden.
|
Hidden and
|
Two outputs.
| false |
QMSum_194
|
Twelve times nine inputs , and a hundred , uh , hidden.
|
Hidden and
|
Two outputs.
|
two outputs.
| false |
QMSum_194
|
Hidden and
|
Two outputs.
|
two outputs.
|
Two outputs. OK. So I guess about eleven thousand parameters , which actually shouldn't be a problem , even in in small phones. Yeah.
| false |
QMSum_194
|
Two outputs.
|
two outputs.
|
Two outputs. OK. So I guess about eleven thousand parameters , which actually shouldn't be a problem , even in in small phones. Yeah.
|
Mm - hmm.
| false |
QMSum_194
|
two outputs.
|
Two outputs. OK. So I guess about eleven thousand parameters , which actually shouldn't be a problem , even in in small phones. Yeah.
|
Mm - hmm.
|
So , I 'm I 'm s so what is different between this and and what you
| false |
QMSum_194
|
Two outputs. OK. So I guess about eleven thousand parameters , which actually shouldn't be a problem , even in in small phones. Yeah.
|
Mm - hmm.
|
So , I 'm I 'm s so what is different between this and and what you
|
It should be OK. So the previous syst It 's based on the system that has a fifty - three point sixty - six percent improvement. It 's the same system. The only thing that changed is the n a p eh a es the estimation of the silence probabilities.
| false |
QMSum_194
|
Mm - hmm.
|
So , I 'm I 'm s so what is different between this and and what you
|
It should be OK. So the previous syst It 's based on the system that has a fifty - three point sixty - six percent improvement. It 's the same system. The only thing that changed is the n a p eh a es the estimation of the silence probabilities.
|
Ah. OK.
| false |
QMSum_194
|
So , I 'm I 'm s so what is different between this and and what you
|
It should be OK. So the previous syst It 's based on the system that has a fifty - three point sixty - six percent improvement. It 's the same system. The only thing that changed is the n a p eh a es the estimation of the silence probabilities.
|
Ah. OK.
|
Which now is based on , uh , cleaned features.
| false |
QMSum_194
|
It should be OK. So the previous syst It 's based on the system that has a fifty - three point sixty - six percent improvement. It 's the same system. The only thing that changed is the n a p eh a es the estimation of the silence probabilities.
|
Ah. OK.
|
Which now is based on , uh , cleaned features.
|
And , it 's a l it 's a lot better.
| false |
QMSum_194
|
Ah. OK.
|
Which now is based on , uh , cleaned features.
|
And , it 's a l it 's a lot better.
|
Wow.
| false |
QMSum_194
|
Which now is based on , uh , cleaned features.
|
And , it 's a l it 's a lot better.
|
Wow.
|
Yeah.
| false |
QMSum_194
|
And , it 's a l it 's a lot better.
|
Wow.
|
Yeah.
|
That 's great.
| false |
QMSum_194
|
Wow.
|
Yeah.
|
That 's great.
|
Um So it 's it 's not bad , but the problem is still that the latency is too large.
| false |
QMSum_194
|
Yeah.
|
That 's great.
|
Um So it 's it 's not bad , but the problem is still that the latency is too large.
|
What 's the latency ?
| false |
QMSum_194
|
That 's great.
|
Um So it 's it 's not bad , but the problem is still that the latency is too large.
|
What 's the latency ?
|
Because um the the latency of the VAD is two hundred and twenty milliseconds. And , uh , the VAD is used uh , i for on - line normalization , and it 's used before the delta computation. So if you add these components it goes t to a hundred and seventy , right ?
| false |
QMSum_194
|
Um So it 's it 's not bad , but the problem is still that the latency is too large.
|
What 's the latency ?
|
Because um the the latency of the VAD is two hundred and twenty milliseconds. And , uh , the VAD is used uh , i for on - line normalization , and it 's used before the delta computation. So if you add these components it goes t to a hundred and seventy , right ?
|
I I 'm confused. You started off with two - twenty and you ended up with one - seventy ?
| false |
QMSum_194
|
What 's the latency ?
|
Because um the the latency of the VAD is two hundred and twenty milliseconds. And , uh , the VAD is used uh , i for on - line normalization , and it 's used before the delta computation. So if you add these components it goes t to a hundred and seventy , right ?
|
I I 'm confused. You started off with two - twenty and you ended up with one - seventy ?
|
With two an two hundred and seventy.
| false |
QMSum_194
|
Because um the the latency of the VAD is two hundred and twenty milliseconds. And , uh , the VAD is used uh , i for on - line normalization , and it 's used before the delta computation. So if you add these components it goes t to a hundred and seventy , right ?
|
I I 'm confused. You started off with two - twenty and you ended up with one - seventy ?
|
With two an two hundred and seventy.
|
Two - seventy.
| false |
QMSum_194
|
I I 'm confused. You started off with two - twenty and you ended up with one - seventy ?
|
With two an two hundred and seventy.
|
Two - seventy.
|
If Yeah , if you add the c delta comp delta computation
| false |
QMSum_194
|
With two an two hundred and seventy.
|
Two - seventy.
|
If Yeah , if you add the c delta comp delta computation
|
Oh.
| false |
QMSum_194
|
Two - seventy.
|
If Yeah , if you add the c delta comp delta computation
|
Oh.
|
which is done afterwards. Um
| false |
QMSum_194
|
If Yeah , if you add the c delta comp delta computation
|
Oh.
|
which is done afterwards. Um
|
So it 's two - twenty. I the is this are these twenty - millisecond frames ? Is that why ? Is it after downsampling ? or
| false |
QMSum_194
|
Oh.
|
which is done afterwards. Um
|
So it 's two - twenty. I the is this are these twenty - millisecond frames ? Is that why ? Is it after downsampling ? or
|
The two - twenty is one hundred milliseconds for the um No , it 's forty milliseconds for t for the , uh , uh , cleaning of the speech. Um then there is , um , the neural network which use nine frames. So it adds forty milliseconds.
| false |
QMSum_194
|
which is done afterwards. Um
|
So it 's two - twenty. I the is this are these twenty - millisecond frames ? Is that why ? Is it after downsampling ? or
|
The two - twenty is one hundred milliseconds for the um No , it 's forty milliseconds for t for the , uh , uh , cleaning of the speech. Um then there is , um , the neural network which use nine frames. So it adds forty milliseconds.
|
a OK.
| false |
QMSum_194
|
So it 's two - twenty. I the is this are these twenty - millisecond frames ? Is that why ? Is it after downsampling ? or
|
The two - twenty is one hundred milliseconds for the um No , it 's forty milliseconds for t for the , uh , uh , cleaning of the speech. Um then there is , um , the neural network which use nine frames. So it adds forty milliseconds.
|
a OK.
|
Um , after that , um , you have the um , filtering of the silence probabilities. Which is a million filter it , and it creates a one hundred milliseconds delay. So , um
| false |
QMSum_194
|
The two - twenty is one hundred milliseconds for the um No , it 's forty milliseconds for t for the , uh , uh , cleaning of the speech. Um then there is , um , the neural network which use nine frames. So it adds forty milliseconds.
|
a OK.
|
Um , after that , um , you have the um , filtering of the silence probabilities. Which is a million filter it , and it creates a one hundred milliseconds delay. So , um
|
Plus there is a delta at the input.
| false |
QMSum_194
|
a OK.
|
Um , after that , um , you have the um , filtering of the silence probabilities. Which is a million filter it , and it creates a one hundred milliseconds delay. So , um
|
Plus there is a delta at the input.
|
Yeah , and there is the delta at the input which is ,
| false |
QMSum_194
|
Um , after that , um , you have the um , filtering of the silence probabilities. Which is a million filter it , and it creates a one hundred milliseconds delay. So , um
|
Plus there is a delta at the input.
|
Yeah , and there is the delta at the input which is ,
|
One hundred milliseconds for smoothing.
| false |
QMSum_194
|
Plus there is a delta at the input.
|
Yeah , and there is the delta at the input which is ,
|
One hundred milliseconds for smoothing.
|
um So it 's @ @
| false |
QMSum_194
|
Yeah , and there is the delta at the input which is ,
|
One hundred milliseconds for smoothing.
|
um So it 's @ @
|
Uh , median.
| false |
QMSum_194
|
One hundred milliseconds for smoothing.
|
um So it 's @ @
|
Uh , median.
|
It 's like forty plus forty plus
| false |
QMSum_194
|
um So it 's @ @
|
Uh , median.
|
It 's like forty plus forty plus
|
And then forty
| false |
QMSum_194
|
Uh , median.
|
It 's like forty plus forty plus
|
And then forty
|
Mmm. Forty This forty plus twenty , plus one hundred.
| false |
QMSum_194
|
It 's like forty plus forty plus
|
And then forty
|
Mmm. Forty This forty plus twenty , plus one hundred.
|
forty p
| false |
QMSum_194
|
And then forty
|
Mmm. Forty This forty plus twenty , plus one hundred.
|
forty p
|
Uh
| false |
QMSum_194
|
Mmm. Forty This forty plus twenty , plus one hundred.
|
forty p
|
Uh
|
So it 's two hundred actually.
| false |
QMSum_194
|
forty p
|
Uh
|
So it 's two hundred actually.
|
Yeah , there are twenty that comes from There is ten that comes from the LDA filters also. Right ?
| false |
QMSum_194
|
Uh
|
So it 's two hundred actually.
|
Yeah , there are twenty that comes from There is ten that comes from the LDA filters also. Right ?
|
Oh , OK.
| false |
QMSum_194
|
So it 's two hundred actually.
|
Yeah , there are twenty that comes from There is ten that comes from the LDA filters also. Right ?
|
Oh , OK.
|
Uh , so it 's two hundred and ten , yeah.
| false |
QMSum_194
|
Yeah , there are twenty that comes from There is ten that comes from the LDA filters also. Right ?
|
Oh , OK.
|
Uh , so it 's two hundred and ten , yeah.
|
If you are using
| false |
QMSum_194
|
Oh , OK.
|
Uh , so it 's two hundred and ten , yeah.
|
If you are using
|
Uh
| false |
QMSum_194
|
Uh , so it 's two hundred and ten , yeah.
|
If you are using
|
Uh
|
Plus the frame ,
| false |
QMSum_194
|
If you are using
|
Uh
|
Plus the frame ,
|
t If you are using three frames
| false |
QMSum_194
|
Uh
|
Plus the frame ,
|
t If you are using three frames
|
so it 's two - twenty.
| false |
QMSum_194
|
Plus the frame ,
|
t If you are using three frames
|
so it 's two - twenty.
|
If you are phrasing f using three frames , it is thirty here for delta.
| false |
QMSum_194
|
t If you are using three frames
|
so it 's two - twenty.
|
If you are phrasing f using three frames , it is thirty here for delta.
|
Yeah , I think it 's it 's five frames , but.
| false |
QMSum_194
|
so it 's two - twenty.
|
If you are phrasing f using three frames , it is thirty here for delta.
|
Yeah , I think it 's it 's five frames , but.
|
So five frames , that 's twenty. OK , so it 's who un two hundred and ten.
| false |
QMSum_194
|
If you are phrasing f using three frames , it is thirty here for delta.
|
Yeah , I think it 's it 's five frames , but.
|
So five frames , that 's twenty. OK , so it 's who un two hundred and ten.
|
Uh , p Wait a minute. It 's forty forty for the for the cleaning of the speech ,
| false |
QMSum_194
|
Yeah , I think it 's it 's five frames , but.
|
So five frames , that 's twenty. OK , so it 's who un two hundred and ten.
|
Uh , p Wait a minute. It 's forty forty for the for the cleaning of the speech ,
|
So. Forty cleaning.
| false |
QMSum_194
|
So five frames , that 's twenty. OK , so it 's who un two hundred and ten.
|
Uh , p Wait a minute. It 's forty forty for the for the cleaning of the speech ,
|
So. Forty cleaning.
|
forty for the I N ANN , a hundred for the smoothing.
| false |
QMSum_194
|
Uh , p Wait a minute. It 's forty forty for the for the cleaning of the speech ,
|
So. Forty cleaning.
|
forty for the I N ANN , a hundred for the smoothing.
|
Yeah.
| false |
QMSum_194
|
So. Forty cleaning.
|
forty for the I N ANN , a hundred for the smoothing.
|
Yeah.
|
Well , but at ten ,
| false |
QMSum_194
|
forty for the I N ANN , a hundred for the smoothing.
|
Yeah.
|
Well , but at ten ,
|
Twenty for the delta.
| false |
QMSum_194
|
Yeah.
|
Well , but at ten ,
|
Twenty for the delta.
|
Twenty for delta.
| false |
QMSum_194
|
Well , but at ten ,
|
Twenty for the delta.
|
Twenty for delta.
|
At th At the input. I mean , that 's at the input to the net.
| false |
QMSum_194
|
Twenty for the delta.
|
Twenty for delta.
|
At th At the input. I mean , that 's at the input to the net.
|
Yeah.
| false |
QMSum_194
|
Twenty for delta.
|
At th At the input. I mean , that 's at the input to the net.
|
Yeah.
|
Delta at input to net ?
| false |
QMSum_194
|
At th At the input. I mean , that 's at the input to the net.
|
Yeah.
|
Delta at input to net ?
|
And there i
| false |
QMSum_194
|
Yeah.
|
Delta at input to net ?
|
And there i
|
Yeah.
| false |
QMSum_194
|
Delta at input to net ?
|
And there i
|
Yeah.
|
Yeah. So it 's like s five , six cepstrum plus delta at nine nine frames of
| false |
QMSum_194
|
And there i
|
Yeah.
|
Yeah. So it 's like s five , six cepstrum plus delta at nine nine frames of
|
And then ten milliseconds for
| false |
QMSum_194
|
Yeah.
|
Yeah. So it 's like s five , six cepstrum plus delta at nine nine frames of
|
And then ten milliseconds for
|
Fi - There 's an LDA filter.
| false |
QMSum_194
|
Yeah. So it 's like s five , six cepstrum plus delta at nine nine frames of
|
And then ten milliseconds for
|
Fi - There 's an LDA filter.
|
ten milliseconds for LDA filter , and t and ten another ten milliseconds you said for the frame ?
| false |
QMSum_194
|
And then ten milliseconds for
|
Fi - There 's an LDA filter.
|
ten milliseconds for LDA filter , and t and ten another ten milliseconds you said for the frame ?
|
For the frame I guess. I computed two - twenty Yeah , well , it 's I guess it 's for the fr the
| false |
QMSum_194
|
Fi - There 's an LDA filter.
|
ten milliseconds for LDA filter , and t and ten another ten milliseconds you said for the frame ?
|
For the frame I guess. I computed two - twenty Yeah , well , it 's I guess it 's for the fr the
|
OK. And then there 's delta besides that ?
| false |
QMSum_194
|
ten milliseconds for LDA filter , and t and ten another ten milliseconds you said for the frame ?
|
For the frame I guess. I computed two - twenty Yeah , well , it 's I guess it 's for the fr the
|
OK. And then there 's delta besides that ?
|
So this is the features that are used by our network and then afterwards , you have to compute the delta on the , uh , main feature stream ,
| false |
QMSum_194
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.