I agree that FFT is not the easiest tool to use for the job. If I were trying to solve this problem, I'd use autocorrelation. But it still sounds very challen

I agree that FFT is not the easiest tool to use for the job. If I were trying to solve this problem, I'd use autocorrelation. But it still sounds very challenging: there's multiple sources of periodic frequency change both in recordings and in playback mechanisms.