Hop time differential issue

[Gordo851]

Hi there!

Love the tech, this project has been incredibly useful in my work analysis conversational data however I recently came across an issue where I noticed that when I extracted features there were fewer returned than I expected. Looking at the closed issues on git hub I realised this was because of the adjustment you make in the background where you find the number of samples to the closest power of 2. This meant that in the background my hop length was adjusted from 10ms to around 11.6ms. This is less than ideal as the papers I've submitted thus far are now misrepresenting what I've done (fortunately none have gone to print and so this is fixable).

I would ask either that you provide the functionality to turn off this adjustment and keep the hop-value to what the user sets it to or you adjust the documentation to make clear what the hop-length parameter does and how surfboard interprets/adjusts it.

For anyone with a similar issue reading this a quick work around is to segment your data yourself by taking slices of the audio and having surfboard extract features from whatever window size you want yourself and then moving the window manually however this is very slow.