使用 wavio Python 3.5 读取 24 位 WAV

Question

I am using wavio by WarrenWeckesser as I need to read 24-bit wav files in python. The wav files I have are produced by some instrumentation and I am trying to get the raw values without any normalisation or scaling.

In the wavio module the code that does the work is this:

    if sampwidth == 3:
        a = _np.empty((num_samples, nchannels, 4), dtype=_np.uint8)
        raw_bytes = _np.fromstring(data, dtype=_np.uint8)
        a[:, :, :sampwidth] = raw_bytes.reshape(-1, nchannels, sampwidth)
        a[:, :, sampwidth:] = (a[:, :, sampwidth - 1:sampwidth] >> 7) * 255
        result = a.view('<i4').reshape(a.shape[:-1]

Can someone explain what it is actually doing (I am a relatively new to numpy and array slicing). I understand most of it but I don't understand what is going on here:

    a[:, :, sampwidth:] = (a[:, :, sampwidth - 1:sampwidth] >> 7) * 255

In my case it does the transform from 24 to 32 bit but I can't work out whether it is scaling the data, or simply padding it out without changing any raw values.

Answer 1

The shape of a is (num_samples, nchannels, 4) and sampwidth == 3 , so that line is the same as

a[:, :, 3:] = (a[:, :, 2:3] >> 7) * 255

which is the same as

a[:, :, 3] = (a[:, :, 2] >> 7) * 255

we could devectorize the outer two loops:

for i in range(num_samples):
    for j in range(nchannels):
        a[i, j, 3] = (a[i, j, 2] >> 7) * 255

The dtype of a is _np.uint8 , so a[...] >> 7 can only give out 0 when the value is <128, or 1 when it is ≥128, so the above becomes:

for i in range(num_samples):
    for j in range(nchannels):
        v = a[i, j, 2]
        a[i, j, 3] = 255 if v >= 128 else 0

If the data are 24-bit little-endian integers, this is equivalent to doing a sign-extension into 32-bit.