Add target dtype argument to save function for sox backend
#1204
Conversation
torchaudio/backend/sox_io_backend.py
Outdated
| ``dtype`` parameter is only effective for ``float32`` Tensor. | ||
| """ | ||
| if src.dtype == torch.float32 and dtype == None: | ||
| warnings.warn('`dtype` default value will be changed to `int16` in 0.9 release') |
There was a problem hiding this comment.
Can you add an instruction for suppress this warning? Many users want to be able to suppress warnings.
Something like "provide dtype argument to suppress this warning."
torchaudio/csrc/sox/io.cpp
Outdated
| throw std::runtime_error("dtype conversion only supported for float32 tensors"); | ||
| } | ||
| const auto tgt_dtype = (tensor.dtype() == torch::kFloat32 && dtype.has_value()) ? | ||
| get_dtype_from_str(dtype.value().c_str()) : tensor.dtype(); |
There was a problem hiding this comment.
Is the call c_str() necessary? It looks like the resulting pointer will be implicitly converted to std::string again.
| sox_io_backend.save(path, data, 8000, dtype=dtype) | ||
| found = load_wav(path, normalize=False)[0] | ||
| self.assertEqual(found.dtype, getattr(torch, dtype)) | ||
| assert(torch.all(found >= range[0]) and torch.all(found <= range[1])) |
There was a problem hiding this comment.
This range assertion does not look quite right. In case of int32 conversion, it will still pass even if the conversion does not happen.
How about performing exact comparison with Tensors that are small enough to catch the element wise parity?
@parameterized.expand([
('float32', torch.Tensor([-1.0, -0.5, 0, 0.5, 1.0]).to(torch.float32)),
('int32', torch.Tensor([-2147483648, ..., 0, ..., 2147483647]).to(torch.int32)),
...
])
def test_dtype_conversion(self, dtype, expected):
data = torch.Tensor([-1.0, -0.5, 0, 0.5, 1.0]).to(torch.dtype32)
...
self.assertEqual(found, expected)
torchaudio/backend/sox_io_backend.py
Outdated
| ``dtype=None`` means no conversion is performed. | ||
| ``dtype`` parameter is only effective for ``float32`` Tensor. | ||
| """ | ||
| if src.dtype == torch.float32 and dtype == None: |
There was a problem hiding this comment.
The save function has to be compatible with TorchScript compiler. Check out "Type Refinement" section to resolve None case.
https://pytorch.org/docs/stable/jit_language_reference.html#optional-type-refinement
I am not sure how TorchScript compiler processes warnings.warn. If the torchscript_compatibility test fails, you can move this warnings to _save function.
torchaudio/backend/sox_io_backend.py
Outdated
| """ | ||
| if src.dtype == torch.float32 and dtype is None: | ||
| warnings.warn( | ||
| '`dtype` default value will be changed to `int16` in 0.9 release' |
There was a problem hiding this comment.
| '`dtype` default value will be changed to `int16` in 0.9 release' | |
| '`dtype` default value will be changed to `int16` in 0.9 release. ' |
torchaudio/backend/sox_io_backend.py
Outdated
| if src.dtype == torch.float32 and dtype is None: | ||
| warnings.warn( | ||
| '`dtype` default value will be changed to `int16` in 0.9 release' | ||
| 'provide `dtype argument to suppress this warning' |
There was a problem hiding this comment.
| 'provide `dtype argument to suppress this warning' | |
| 'Specify `dtype` to suppress this warning.' |
|
Thanks! |
* add new speech tutorial. * update with a few parameter tuned. model takes less than 10 min to run now. * feedback. * improve GPU performance. add interactive demo at the end. * feedback.
3 participants
issue: #1197