Why does changing the format of an audio file from int16 to float produce jitter?

Question

I read an audio file (an instrumental music file that is mono and 16000) with wav.read (read function from scipy.io.wavfile.) This function gives me a data file in int16 format. When I change the type to float and write it again I hear a lot of jitter. Why does that happen?

Here is my script in python 3.6. This is the music file.

import scipy.io.wavfile as wav
from scipy.io.wavfile import write
rate, music = wav.read("music.wav")
music = np.expand_dims(music, -1)
music_float = music.astype('float')
write("float_music", rate, music_float)

Answer 1

The reason for this is that wavfile.write supports different number ranges for different data types.

The Valid ranges are

WAV format	Min	Max	NumPy dtype
32-bit floating-point	-1.0	+1.0	float32
32-bit PCM	-2147483648	+2147483647	int32
16-bit PCM	-32768	+32767	int16
8-bit PCM	0	255	uint8

(source https://docs.scipy.org/doc/scipy/reference/generated/scipy.io.wavfile.write.html)

This is why Peters solution works.

Answer 2

As Marcus and you say, try normalizing the float version before writing.

Changing

write("float_music", rate, music_float)

to

write("unit_flow_music.wav", rate, music_float/max(np.abs(music_float)))

seems to do the trick for me.