Online Participants Needed for Research Experiment

[keelan_hough]

Hello, we are students studying at the University of Huddersfield, and we are conducting an experiment focused on comparing consumer grade separation algorithms. If we could please have 10-15 minutes of your time to complete an online survey that'd be appreciated massively.
The test will have you listen to 6 stimuli, which are vocals of a segment of a track that has been run through the different algorithms and you need to rate them on a scale. There are 3 tracks used, and 3 tests per track, each test having the track mixed slightly differently to see if things like vocal panning or hyper-compression would affect the algorithm's ability to separate vocals of the track.
We ask that you please use headphones during the experiment.
The link for the test is: https://golisten.ucd.ie/task/mushra-tes ... 7eae0082be
Thanks in advance

[James Perrett]

I may be missing something but I've stumbled at the first test. The first sample sounds like an AM radio of a complete track, the next 5 sound like separated vocals while the reference still sounds artificial - as if it has been excessively processed already.

How can we evaluate the separation algorithms if the source already sounds unnatural? Did you not notice the processing artefacts before designing this test?

[keelan_hough]

The reference is the vocals from a Cambridge multitracks package, which did come already processed which is why it sounds unnatural. The rating part just boils down to how you think they sound in terms of quality and similarity to the reference. If they sound worse than or just different then rate them lower, please.

[RichardT]

Do you have time to rethink?

Using something that sounds processed at the start leaves you open to getting confused responses from people here, and it’s really not the best in terms of experimental design, where I would have thought the best way to identify processing anomalies is to start with as clean an input source as possible.

I’m sure Mike Senior has some clean mixes somewhere.

[Eddy Deegan]

I bailed out half way through as well, that first test is very strange and I didn't feel it worth completing.

[James Perrett]

The reference is the vocals from a Cambridge multitracks package,

Huddersfield University has its own recording studios.

https://research.hud.ac.uk/music/facilities/studios/

There should be no need to use someone else's work when you should already have (or can easily create) the necessary recordings on campus.

I am also surprised that your survey doesn't give more information about how the data will be used. Previous surveys from your university that I have seen have included full details about how the data will be used.

[ilikebarmouth]

That went well.

[amanise]

[The Elf]

As above. No idea which anomalies I'm supposed to accept and ignore, and which I'm supposed to critically evaluate.