Hi there !

I've spend a some time figuring it out but wav file read is normalized...
To me (as an acoustic engineer) it makes absolutely no sense :)
Especially since there are some calibrate_to and calibrate_with functions.
Am I missing something ? Why would you normalize on import ?

I ended up with the following code to read from the wave file :

import acoustics
import wave
from scipy.io import wavfile
file_path = "whatever.wav"
with wave.open(file_path, 'rb') as wav:
    sampwidth = wav.getsampwidth()
fs, data = wavfile.read(file_path) 
data = data / 2**(8*sampwidth-1)
signal= acoustics.Signal(data, fs=fs)

This works with S16_LE, S32_LE, S24_3LE formats (using the ALSA labelling) but would fail with S24_LE or FLOAT_LE .
It would be nice to have any wav red as a float array in the end.

I would be happy to contribute, I just want to make sure I didn't misunderstand an intentional design choice.