ENH: improve dtype check in wavfile.write

[JuanFMontesinos]

What does this implement/fix?

scipy.io.wavfile.write has a wrong data type check. It checks a numpy array to be "float" or "integer" in a generic way. However float16 is not supported by any codec but accepted by the function, leading to a silent error.
Likewise, exotic data types such as float128, int128 and so on are neither supported.

Additional information

While it's complicated to find what dtypes are exactly supported, I rely on ffmpeg formats as they are a superset of scipy's ones.
int64 is also supported despite not appearing.

From your references

[tylerjereddy]

In case the team wants a regression test I noticed that this fails on main and passes here:

--- a/scipy/io/tests/test_wavfile.py
+++ b/scipy/io/tests/test_wavfile.py
@@ -414,3 +414,15 @@ def test_write_roundtrip(realfile, mmap, rate, channels, dt_str, tmpdir):
         # in PyPy; since the filename gets reused in this test, clean this up
         break_cycles()
         break_cycles()
+
+
+@pytest.mark.parametrize("dtype", [
+    np.float16,
+    ])
+def test_wavfile_dtype_unsupported(tmpdir, dtype):
+    tmpfile = str(tmpdir.join('temp.wav'))
+    rng = np.random.default_rng(1234)
+    data = rng.random((100, 5)).astype(dtype)
+    rate = 8000
+    with pytest.raises(ValueError, match="Unsupported"):
+        wavfile.write(tmpfile, rate, data)

[tylerjereddy]

I know int8 was already allowed in the old check, but I'm wondering about it specifically because in scipy/io/tests/test_wavfile.py above test_write_roundtrip I see signed 8-bit integer PCM is not allowed and a similar statement in the docs indicating that Note that 8-bit PCM is unsigned.

[JuanFMontesinos]

Well, I don't know the intrinsecs but the docs only show common data types, not all data types.
The second table is extracted from the references provided in the docs:
IBM Corporation and Microsoft Corporation, “Multimedia Programming Interface and Data Specifications 1.0”, section “Data Format of the Samples”, August 1991 http://www.tactilemedia.com/info/MCI_Control_Info.html

I need to recheck but I think I did a quick test and ffprobe was accepting the output for int8.

[JuanFMontesinos]

You are right, according to ffprobe info it seems it's processed as a uint8. Fixed at 33b8a51

[j-bowhay]

@JuanFMontesinos could you add the suggested regression test, please?

[JuanFMontesinos]

@j-bowhay

Done at 37f0d17! Although tests are failing due to an import error out of my control.

[lucascolley]

Hey @JuanFMontesinos , would you mind re-triggering CI here? Pushing an empty commit will do it.

[JuanFMontesinos]

Hey @JuanFMontesinos , would you mind re-triggering CI here? Pushing an empty commit will do it.

Done!

[lucascolley]

Okay, the docs build should pass if you merge main into this branch. Then a maintainer can approve the GHA workflows.

[lucascolley]

CI is happy, the change is simple enough and the regression test is there, so pulling this in. Thanks for your first contribution to SciPy @JuanFMontesinos ! And thanks for the test @tylerjereddy

[JuanFMontesinos]

@lucascolley @tylerjereddy Thank you very much and thanks all the developers for such a nice work with SciPy.