4 ms·

It is, but I'm sure the audible artifacts introduced by a pitch shifter would outweigh any perceptual effects due to the change in tuning. It has to be one of t

by cannam 5y ago

It is, but I'm sure the audible artifacts introduced by a pitch shifter would outweigh any perceptual effects due to the change in tuning. It has to be one of the two - either the lower-frequency music is fractionally slower or it has been modified in a non-transparent way - I don't think the article tells us which. There seem to be other things the article doesn't tell us, such as how the authors verified that the music they used had been tuned to A=440 in the first place (since it's far from universal) or why, given an infinite number of possible tuning frequencies, a great many of which have been chosen by musicians at some point, and given no obvious physical basis for this specific choice, they went for A=432 instead of any other value below 440. [edit: removed warranted but uncivil snark]

insickness 5y ago
You can transparently slow music without changing the pitch or pitch music without changing the tempo far more than 2%, which was what was done here. This technology has been in use by DJs spinning CDs since the 90s and is also used in music production. Depending on the source material, you can actually double the pitch or tempo with no artifacts.
- cannam 5y ago
  I agree with you that you can often do this in a way that is still pleasant to listen to. But timestretching/pitch shifting is very far from being transparent in the way that a decent resampler is. Even a small shift will introduce some audible artifacts in most music. Will they be more or less apparent to the listener than the difference between recordings tuned to A=440 or 432 would be? I don't know. Depends on the music and the listener, I imagine. That would (be one of the many, many unstated things that would) matter to a paper like this. (For what it's worth, I wrote and maintain a widely-used timestretcher library. Possibly this means I am more aware of the limitations than the strengths of such technology)
- marcan_42 5y ago
  There is no way to do this with no artifacts; it's just fundamental to how audio works. We have pretty good algorithms for it, but in the end you're doing psychoacoustics to try to make it sound good. There's no mathematical model for doing this "perfectly" like there is for resampling. It's not hard to hear the artifacts in DJ tempo stretch algorithms if you listen closely, especially when they are applied to a full mix (as opposed to single instruments). Of course they're small these days, but when you're trying to measure a subtle effect, they're too much of a confounding variable.

by cannam 5y ago

insickness 5y ago

You can transparently slow music without changing the pitch or pitch music without changing the tempo far more than 2%, which was what was done here. This technology has been in use by DJs spinning CDs since the 90s and is also used in music production. Depending on the source material, you can actually double the pitch or tempo with no artifacts.

cannam 5y ago
I agree with you that you can often do this in a way that is still pleasant to listen to. But timestretching/pitch shifting is very far from being transparent in the way that a decent resampler is. Even a small shift will introduce some audible artifacts in most music. Will they be more or less apparent to the listener than the difference between recordings tuned to A=440 or 432 would be? I don't know. Depends on the music and the listener, I imagine. That would (be one of the many, many unstated things that would) matter to a paper like this. (For what it's worth, I wrote and maintain a widely-used timestretcher library. Possibly this means I am more aware of the limitations than the strengths of such technology)
marcan_42 5y ago
There is no way to do this with no artifacts; it's just fundamental to how audio works. We have pretty good algorithms for it, but in the end you're doing psychoacoustics to try to make it sound good. There's no mathematical model for doing this "perfectly" like there is for resampling. It's not hard to hear the artifacts in DJ tempo stretch algorithms if you listen closely, especially when they are applied to a full mix (as opposed to single instruments). Of course they're small these days, but when you're trying to measure a subtle effect, they're too much of a confounding variable.