Stereo to Mono wav in Python

Question

I am loading a wav with the scipy method wavefile.read() which gives me the samplerate and the audiodata

I know that this audio data if stereo is stored as a multi-dimensional array such as

audiodata[[left right]
          [left right]
          ...
          [left right]]

I am then using this method to create a new array of mono audio data by taking (right+left)/2

def stereoToMono(audiodata)
    newaudiodata = []

    for i in range(len(audiodata)):
        d = (audiodata[i][0] + audiodata[i][1])/2
        newaudiodata.append(d)

    return np.array(newaudiodata, dtype='int16')

and then i write this to file using

wavfile.write(newfilename, sr, newaudiodata)

This is producing a Mono wav file, however the sound is dirty and has clickd etc throughout

what am I doing wrong?

Solution 1

First, what is the datatype of audiodata? I assume it's some fixed-width integer format and you therefore get overflow. If you convert it to a floating point format before processing, it will work fine:

audiodata = audiodata.astype(float)

Second, don't write your Python code element by element; vectorize it:

d = (audiodata[:,0] + audiodata[:,1]) / 2

or even better

d = audiodata.sum(axis=1) / 2

This will be vastly faster than the element-by-element loop you wrote.

Solution 2

turns out, all i had to change was

(right+left)/2

to

(right/2) + (left/2)

Comments

[WinEunuuchs2Unix]

Shouldn't you accept the answer you went with?

[b-rad15]

@WinEunuuchs2Unix not when the accepted answer is better. Converting to a float saves some loss when the integers are odd

[Peter Julian]

Why not audiodata.mean(axis=1)?