topic_shift
bool 2
classes | utterance
stringlengths 1
7.9k
| session_id
stringlengths 7
14
|
|---|---|---|
false | Yeah. | QMSum_120 |
false | The filters roughly act like a , um a , uh a an overall resonant you know , f some resonances and so forth that th that 's processing excitation. | QMSum_120 |
false | Here. | QMSum_120 |
false | They should be more close. | QMSum_120 |
false | Ah , no. This is this ? More close. Is this ? And this. | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | Yeah. | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | So they are this is there is less difference. | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | So if you look at the spectral envelope , just the very smooth properties of it , you get something closer to that. | QMSum_120 |
false | This is less it 's less robust. | QMSum_120 |
false | Less robust. Yeah. | QMSum_120 |
false | Oh , yeah. | QMSum_120 |
false | And the notion is if you have the full spectrum , with all the little nitty - gritty details , that that has the effect of both , | QMSum_120 |
false | Yeah. | QMSum_120 |
false | and it would be a multiplication in in frequency domain | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | so that would be like an addition in log power spectrum domain. | QMSum_120 |
false | Mm - hmm. Mm - hmm. | QMSum_120 |
false | And so this is saying , well , if you really do have that sort of vocal tract envelope , and you subtract that off , what you get is the excitation. And I call that lies because you don't really have that , you just have some kind of signal - processing trickery to get something that 's kind of smooth. It 's not really what 's happening in the vocal tract | QMSum_120 |
false | Yeah. | QMSum_120 |
false | so you 're not really getting the vocal excitation. | QMSum_120 |
false | Right. | QMSum_120 |
false | That 's why I was going to the why I was referring to it in a more a more , uh , uh , conservative way , when I was saying " well , it 's yeah , it 's the excitation ". But it 's not really the excitation. It 's whatever it is that 's different between | QMSum_120 |
false | Oh. This moved in the | QMSum_120 |
false | So so , stand standing back from that , you sort of say there 's this very detailed representation. | QMSum_120 |
false | Yeah. | QMSum_120 |
false | You go to a smooth representation. | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | You go to a smooth representation cuz this typically generalizes better. | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | Um , but whenever you smooth you lose something , so the question is have you lost something you can you use ? | QMSum_120 |
false | Right. | QMSum_120 |
false | Um , probably you wouldn't want to go to the extreme of just ta saying " OK , our feature set will be the FFT " , cuz we really think we do gain something in robustness from going to something smoother , but maybe there 's something that we missed. | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | So what is it ? | QMSum_120 |
false | Yeah. | QMSum_120 |
false | And then you go back to the intuition that , well , you don't really get the excitation , but you get something related to it. | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | And it and as you can see from those pictures , you do get something that shows some periodicity , uh , in frequency , | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | you know , and and and also in time. | QMSum_120 |
false | Hmm. | QMSum_120 |
false | So | QMSum_120 |
false | That 's that 's really neat. | QMSum_120 |
false | so , | QMSum_120 |
false | So you don't have one for unvoiced picture ? | QMSum_120 |
false | Uh , not here. | QMSum_120 |
false | Oh. | QMSum_120 |
false | No , I have s | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | Yeah. | QMSum_120 |
false | But not here. | QMSum_120 |
false | But presumably you 'll see something that won't have this kind of , uh , uh , uh , regularity in frequency , uh , in the | QMSum_120 |
false | But Yeah. Well. | QMSum_120 |
false | Not here. | QMSum_120 |
false | I would li I would like to see those pictures. | QMSum_120 |
false | Well , so. | QMSum_120 |
false | Yeah. | QMSum_120 |
false | I can't see you now. | QMSum_120 |
false | Yeah. | QMSum_120 |
false | Yeah. | QMSum_120 |
false | Yeah. | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | I don't have. | QMSum_120 |
false | And so you said this is pretty doing this kind of thing is pretty robust to noise ? | QMSum_120 |
false | It seems , yeah. Um , | QMSum_120 |
false | Huh. | QMSum_120 |
false | Pfft. Oops. The mean is different with it , because the the histogram for the the classifica | QMSum_120 |
false | No , no , no. But th the kind of robustness to noise | QMSum_120 |
false | Oh ! | QMSum_120 |
false | So if if you take this frame , uh , from the noisy utterance and the same frame from the clean utterance | QMSum_120 |
false | Hmm. | QMSum_120 |
false | You end up with a similar difference | QMSum_120 |
false | Y y y yeah. We end up with | QMSum_120 |
false | over here ? | QMSum_120 |
false | Yeah. | QMSum_120 |
false | OK. Cool ! | QMSum_120 |
false | I have here the same frame for the clean speech | QMSum_120 |
false | Oh , that 's clean. | QMSum_120 |
false | the same cle | QMSum_120 |
false | Oh , OK | QMSum_120 |
false | But they are a difference. | QMSum_120 |
false | Yeah , that 's | QMSum_120 |
false | Because here the FFT is only with two hundred fifty - six point | QMSum_120 |
false | Oh. | QMSum_120 |
false | and this is with five hundred twelve. | QMSum_120 |
false | Yeah. This is kind of inter interesting also | QMSum_120 |
false | OK. | QMSum_120 |
false | because if we use the standard , uh , frame length of of , like , twenty - five milliseconds , um , what happens is that for low - pitched voiced , because of the frame length , y you don't really have you don't clearly see this periodic structure , | QMSum_120 |
false | Mm - hmm. | QMSum_120 |
false | because of the first lobe of of each each of the harmonics. | QMSum_120 |
false | So this one inclu is a longer Ah. | QMSum_120 |
false | So , this is like yeah , fifty milliseconds or something like that. | QMSum_120 |
false | Fifty millis Yeah. | QMSum_120 |
false | Yeah , but it 's the same frame and | QMSum_120 |
false | Oh , it 's that time - frequency trade - off thing. | QMSum_120 |
false | Yeah. | QMSum_120 |
false | Right ? I see. Yeah. | QMSum_120 |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.