Non-linear speed adjustment in audition

[j7706278]

Does anyone know if it is possible to do a nonlinear speed adjustment in audition? The issue is that I have an old analog recording that is at the wrong speed, but it is a different wrong speed at the beginning then it is at the end, and the recording is about 15 minutes long. It is a very gradual change from beginning to end. I never thought that a 60Hz hum would be useful, but when you plot the spectrogram of the entire recording, you can see that a narrow-band of energy near 60Hz (correspoinding to the hum) linearly changes from beginning to end, so this means that while the recording is accelerating, it is a constant acceleration. Therefore, it should be possible to place two key points, one at the beginning, and one at the end, saying how much speed adjustment should be done and then the speed adjustment should linearly change going from start to end. Is this possible with audition? I know one could segment the recording into windows and do different speed adjustments in each segment, but that would be very time consuming and also inacurate since at the boundary of each segment the adjustment would be less accurate, not to mention transient effects that would probably be audible at those boundaries. I was looking at things like "flex time" in logic pro, and also "elastic audio" in a program called pitch n time, and IRCAM has something called "Transpose/Stretching", but they seem to do a lot more than what I need here, and also I already have an audition licence so would like to be able to use audition if possible. Thanks for any help! 

[SteveG_AudioMasters_]

Gradual speed adjustments generally don't work very well, however you do them - with one exception that I'll come to. The problem is that at any given moment, you are essentially saying that the sample rate's wrong - but only a little bit and because it's a linear variation it's never going to be absolutely correct at that instant anyway - because it's constantly changing. So it's always chasing its own tail.

As I said, there's only one exception, and that's Celemony's Capstan. It's designed to do exactly what you are after (and a bit...) but it's the only software that addresses the fundamental issue I mentioned above. Search for an alternative, and you'll find a lot of frustrated people, because Capstan is expensive... even to rent for a week (200 euros). Buying it? Yeah right - that will cost you a cool 3790 euros!

https://www.celemony.com/en/capstan

Unfortunately there's no way that Audition can get anywhere close to this.

[j7706278]

Thanks very much for the pointer to capstan. That indeed seems very capable.

I don't think, however, that what I'm asking for is technically insurmountable, and it actually should be fairly straightforward mathematically. For example, one possible solution would upsample the signal by a very large amount. I.e. by a factor of 1000 or so, or maybe higher. The resulting sample rate would be high enought that one could in effect treat it as a continuous time signal. Given the high sample rate signal, one could then re-sample the signal at a varying rate, linearly changing the rate from start to finish according to the constant acceleration I mentioned above. Yes, the time points would not perfectly match up the time quanta of the upsampled signal, but with such a high upsampling, the error would probably be imperceptable. It's a program I could write in matlab and/or python fairly easily. I recorded the audio tape at 48kHz 16-bit mono, which for a 15 minute recording takes about 41MB of memory. Upsampling by a factor of 1000 would take about 40GB of memory, which is quite feasible on todays machines to hold all in memory at the same time. Due to the band-limited low-pass nature of the original recording, I almost certainly don't need the 48kHz original sampling rate, so doing the above might even be overkill. However, if there is software to do it (ideally audition), I'd rather not do it myself (but neither pay 200 euros for a one-time task :-). 

thanks again for the tip!