scipy/wavfile.patch

37 lines
1.7 KiB
Diff

commit 09d753f0ae71441906f5cee7a44b2d2b80212082
Author: Nikola Forró <nforro@redhat.com>
Date: Thu Jan 14 14:34:14 2021 +0100
ENH: Support big-endian platforms and big-endian WAVs
PR #12287 added support for reading arbitrary-bit-depth WAVs, but
the code doesn't consider big-endian WAVs, and doesn't work as expected
on big-endian platforms due to the use of native-byte-order data-types.
This change fixes that.
There is also a simple test case that compares euqivalent RIFX
(big-endian) and RIFF (little-endian) files to verify the data read
are the same.
diff --git a/scipy/io/wavfile.py b/scipy/io/wavfile.py
index 9b5845d6b..951f8d201 100644
--- a/scipy/io/wavfile.py
+++ b/scipy/io/wavfile.py
@@ -458,10 +458,13 @@ def _read_data_chunk(fid, format_tag, channels, bit_depth, is_big_endian,
if dtype == 'V1':
# Rearrange raw bytes into smallest compatible numpy dtype
- dt = numpy.int32 if bytes_per_sample == 3 else numpy.int64
- a = numpy.zeros((len(data) // bytes_per_sample, dt().itemsize),
+ dt = f'{fmt}i4' if bytes_per_sample == 3 else f'{fmt}i8'
+ a = numpy.zeros((len(data) // bytes_per_sample, numpy.dtype(dt).itemsize),
dtype='V1')
- a[:, -bytes_per_sample:] = data.reshape((-1, bytes_per_sample))
+ if is_big_endian:
+ a[:, :bytes_per_sample] = data.reshape((-1, bytes_per_sample))
+ else:
+ a[:, -bytes_per_sample:] = data.reshape((-1, bytes_per_sample))
data = a.view(dt).reshape(a.shape[:-1])
else:
if bytes_per_sample in {1, 2, 4, 8}: