How to read a MP3 audio file into a numpy array / save a numpy array to MP3?

Question

Is there a way to read/write a MP3 audio file into/from a numpy array with a similar API to scipy.io.wavfile.read and scipy.io.wavfile.write:

sr, x = wavfile.read('test.wav')
wavfile.write('test2.wav', sr, x)

?

Note: pydub's AudioSegment object doesn't give direct access to a numpy array.

^{PS: I have already read Importing sound files into Python as NumPy arrays (alternatives to audiolab), tried all the answers, including those which requires to Popen ffmpeg and read the content from stdout pipe, etc. I have also read Trying to convert an mp3 file to a Numpy Array, and ffmpeg just hangs, etc., and tried the main answers, but there was no simple solution. After spending hours on this, I'm posting it here with "Answer your own question – share your knowledge, Q&A-style". I have also read How to create a numpy array from a pydub AudioSegment? but this does not easily cover the multi channel case, etc.}

Answer 1

Calling ffmpeg and manually parsing its stdout as suggested in many posts about reading a MP3 is a tedious task (many corner cases because different number of channels are possible, etc.), so here is a working solution using pydub (you need to pip install pydub first).

This code allows to read a MP3 to a numpy array / write a numpy array to a MP3 file with a similar API than scipy.io.wavfile.read/write:

import pydub 
import numpy as np

def read(f, normalized=False):
    """MP3 to numpy array"""
    a = pydub.AudioSegment.from_mp3(f)
    y = np.array(a.get_array_of_samples())
    if a.channels == 2:
        y = y.reshape((-1, 2))
    if normalized:
        return a.frame_rate, np.float32(y) / 2**15
    else:
        return a.frame_rate, y

def write(f, sr, x, normalized=False):
    """numpy array to MP3"""
    channels = 2 if (x.ndim == 2 and x.shape[1] == 2) else 1
    if normalized:  # normalized array - each item should be a float in [-1, 1)
        y = np.int16(x * 2 ** 15)
    else:
        y = np.int16(x)
    song = pydub.AudioSegment(y.tobytes(), frame_rate=sr, sample_width=2, channels=channels)
    song.export(f, format="mp3", bitrate="320k")

Notes:

• It only works for 16-bit files for now (even if 24-bit WAV files are pretty common, I've rarely seen 24-bit MP3 files... Does this exist?)
• normalized=True allows to work with a float array (each item in [-1,1))

Usage example:

sr, x = read('test.mp3')
print(x)

#[[-225  707]
# [-234  782]
# [-205  755]
# ..., 
# [ 303   89]
# [ 337   69]
# [ 274   89]]

write('out2.mp3', sr, x)

Answer 2

You can use audio2numpy library. Install with

pip install audio2numpy

Then, your code would be:

import audio2numpy as a2n
x,sr=a2n.audio_from_file("test.mp3")

For writing, use @Basj 's answer