(no subject)
Dec. 7th, 2005 03:12 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
eiriasI just got a demonstration of the importance of the envelope of a sound. (For the laymen out there, which is probably most of you, "envelope" refers to the shape of the sound over time, including attack, or how it begins, and decay, or how it dies out -- really, everything that happens after the attack is decay.)
I went down to the lab to see how my stimuli sound over the headphones used in real experiments. Answer: exposure and test stimuli don't sound the same, in that exposure sounds are a believable piano imitation and test sounds aren't. Scratch my head. Well, maybe I accidentally set each to different instrument sounds when dumping from MIDI to WAV.
I come up and check. Nope, they're all set on "Acoustic Grand Piano." More head scratching.
I try a different program to listen in -- a sound editor that, incidentally, lets me look at the shape of the waveform. I open exposure and test. I play them repeatedly and listen carefully for what it is, exactly, that's different.
Then I notice the visual difference -- the apparent silence at the beginning of the exposure item that isn't matched by a similar shape in the test item -- and remember that for the sake of convenience, when trying to make all of my test stimulus chords the same length with identical attacks, for each I clipped 600 ms out of the middle of a longer sound file and recreated the attack with a quick upramp, thinking it wouldn't make that big a difference. Well, duh, Sherlock, it makes a difference, in that with a clumsy attack it sounds like a cheap imitation of a piano instead of a pretty good imitation of a piano.
And not only that, but the test stimuli *also* suck because they're taken from quite far in -- by the beginning of the clip, the sounds have had a good 200 ms to decay, and it turns out that the people who made this patch got the attack right but had a harder time making the decay sound natural, chiefly because it's a harder problem. Strike a piano key and hold it down, the sound will eventually die out in this natural, elegant way; strike a keyboard key and hold it down, and the instrument has no choice but to continue in this degenerate and sad fashion, because after all you're telling it to keep holding on, right?
I should have thought of both of these things before, because after all I am not a total sound processing n00b, but I had never come across such a clear demonstration of the fact before today.
I don't know how to solve this, exactly, because the reason I did it the dumb way still holds -- there's more variability in onset time than I want when I dump from MIDI to WAV (being off by even a few milliseconds sucks when you're trying to sync stimuli with brain activity, and also rhythmic irregularities will distract people). But it's good news in the sense that it's (a) not my sound card's fault, which I was sure was the case, because it really is a crappy sound card and (b) not an indication that I need to start the stimulus recording process all over again, because I do have the longer, unprocessed single-chord WAV files, so once I figure out a cleverer solution, it'll just take a few minutes of script-writing to fix.
I went down to the lab to see how my stimuli sound over the headphones used in real experiments. Answer: exposure and test stimuli don't sound the same, in that exposure sounds are a believable piano imitation and test sounds aren't. Scratch my head. Well, maybe I accidentally set each to different instrument sounds when dumping from MIDI to WAV.
I come up and check. Nope, they're all set on "Acoustic Grand Piano." More head scratching.
I try a different program to listen in -- a sound editor that, incidentally, lets me look at the shape of the waveform. I open exposure and test. I play them repeatedly and listen carefully for what it is, exactly, that's different.
Then I notice the visual difference -- the apparent silence at the beginning of the exposure item that isn't matched by a similar shape in the test item -- and remember that for the sake of convenience, when trying to make all of my test stimulus chords the same length with identical attacks, for each I clipped 600 ms out of the middle of a longer sound file and recreated the attack with a quick upramp, thinking it wouldn't make that big a difference. Well, duh, Sherlock, it makes a difference, in that with a clumsy attack it sounds like a cheap imitation of a piano instead of a pretty good imitation of a piano.
And not only that, but the test stimuli *also* suck because they're taken from quite far in -- by the beginning of the clip, the sounds have had a good 200 ms to decay, and it turns out that the people who made this patch got the attack right but had a harder time making the decay sound natural, chiefly because it's a harder problem. Strike a piano key and hold it down, the sound will eventually die out in this natural, elegant way; strike a keyboard key and hold it down, and the instrument has no choice but to continue in this degenerate and sad fashion, because after all you're telling it to keep holding on, right?
I should have thought of both of these things before, because after all I am not a total sound processing n00b, but I had never come across such a clear demonstration of the fact before today.
I don't know how to solve this, exactly, because the reason I did it the dumb way still holds -- there's more variability in onset time than I want when I dump from MIDI to WAV (being off by even a few milliseconds sucks when you're trying to sync stimuli with brain activity, and also rhythmic irregularities will distract people). But it's good news in the sense that it's (a) not my sound card's fault, which I was sure was the case, because it really is a crappy sound card and (b) not an indication that I need to start the stimulus recording process all over again, because I do have the longer, unprocessed single-chord WAV files, so once I figure out a cleverer solution, it'll just take a few minutes of script-writing to fix.
![[identity profile]](https://www.dreamwidth.org/img/silk/identity/openid.png){kind=small}
(no subject)
Date: 2005-12-07 02:09 pm (UTC)(no subject)
Date: 2005-12-07 04:01 pm (UTC)I would love it more if I weren't under a lot of time pressure to get subjects now now now ;). Certainly it's useful to learn stuff about sound processing, but on the other hand ...
WAIT! I figured out how to solve the problem! And it isn't going to take any time to do because I can automate it! OMG OMG SQUEE!
N.B. if you ever do an audio experiment: Adobe Audition? good program, but just a tad frustrating.
(no subject)
Date: 2005-12-07 04:51 pm (UTC)(no subject)
Date: 2005-12-08 07:56 am (UTC)(no subject)
Date: 2005-12-08 03:19 pm (UTC)Out of curiousity (and you may want to elaborate in a new entry for the benefit of all), what does this have to do with babies and pseudo-Finnish words?
(no subject)
Date: 2005-12-08 08:26 pm (UTC)