Closed est31 closed 4 years ago
I scaled the input floats by 65535, the input range in the C example code uses shorts as input so 0..65535
should be the best range. Then it works perfect!
Thanks for that idea! It now works indeed. I'm adding the conversion code to the sample.
Btw, c shorts are in the range of -32767..=32767... so one should rather multiply it by that number. It's working for you because it's only a factor of two difference in the amplitude which the model can obviously handle.
If you do the following:
You get a
car_0dBdn.wav
that sounds basically the same ascar_0dB.wav
. Not sure why that's the case. We are using the builtin model, maybe it is not trained? IDK. cc @jmvalin @jeremyandrews .