Question - About audio input resampling

Hello, according to the util.load method https://github.com/f90/Wave-U-Net/blob/1dd50fdce39edb68c43c85413835a608432070d9/Utils.py#L110 you are loading the audio file through librosa.load but not doing any resampling here, while running a resampling by defaults of sr=22050 with scypy. resample_poly method. I’m using produce_source_estimates where the audio is loaded like

audio, sr = Utils.load(input_path, sr=None, mono=False)

so it’s not using any resampling by defaults. Any reason for that? In my CNN I have a resampling of 12KHz of the input, so can I do apply this resampling?

[f90]

Ah OK, so the classifier is also a Tensorflow model and you want to run it all in one session? In that case you want to keep everything as is, except modify

https://github.com/f90/Wave-U-Net/blob/master/Evaluate.py#L73

where the Wave-U-Net output gets resampled - you can change that so it always gets resampled to 12KHz, and then feed it into your model. Note that the code here actually starts a new TF session each time though - which if you are training your own model seems unnecessary, so you might want to remove that stuff too so both models are in the same persistent session

[loretoparisi]

Thank you for your help! It makes sense to run everything in the same session! 💯