Add TorchScript-able "save" func to sox_io backend #732
Conversation
mthrok
force-pushed
the
save
branch
2 times, most recently
from
fcc7e45 to
5ed66a3
Compare
mthrok
force-pushed
the
save
branch
21 times, most recently
from
e96727e to
c9fd168
Compare
Codecov Report
@@ Coverage Diff @@
## master #732 +/- ##
==========================================
- Coverage 89.23% 89.02% -0.22%
==========================================
Files 32 32
Lines 2517 2532 +15
==========================================
+ Hits 2246 2254 +8
- Misses 271 278 +7
Continue to review full report at Codecov.
|
mthrok
force-pushed
the
save
branch
5 times, most recently
from
2b50660 to
1f49bd9
Compare
mthrok
force-pushed
the
save
branch
17 times, most recently
from
e88aba5 to
b9f8732
Compare
| elif ext in ['ogg', 'vorbis']: | ||
| compression = 3. | ||
| else: | ||
| raise RuntimeError(f'Unsupported file type: "{ext}"') |
There was a problem hiding this comment.
@vincentqb I thought of calling for opening an issue for adding support for other audio formats, but differentiating the valid extensions that are supported by libsox from invalid ones is not trivial and there are many formats that require additional libraries to be installed. Meaning that adding support on our get_signalinfo function may not be enough for users to actually use a new format, depending on the format. So I think we want to be more cautious and not giving high expectations to users casually. Therefore I decided not to include call for opening a PR in the error message.
There was a problem hiding this comment.
Can you remind me how the current backend exposes more backend? From our discussion, the user had to provide extra internal information, right?
There was a problem hiding this comment.
The existing implementation asks users to construct the following data structure, which requires users to know the internal of libsox.
https://fossies.org/dox/sox-14.4.2/structsox__signalinfo__t.html
https://fossies.org/dox/sox-14.4.2/structsox__encodinginfo__t.html
| # note: torchaudio can load large vorbis file, but cannot save large volbis file | ||
| # the following test causes Segmentation fault |
There was a problem hiding this comment.
Is this meant to be fixed eventually?
There was a problem hiding this comment.
Nope I do not think this can be fixed. However, I wanted to leave a record of know limitation.
| elif ext in ['ogg', 'vorbis']: | ||
| compression = 3. | ||
| else: | ||
| raise RuntimeError(f'Unsupported file type: "{ext}"') |
There was a problem hiding this comment.
Can you remind me how the current backend exposes more backend? From our discussion, the user had to provide extra internal information, right?
|
thanks |
* fix loss calculation for RNN * fixes loss for both RNN & Transformer
This is a part of PRs to add new "sox_io" backend. #726 and depends on #718, #728 and #731.
This PR adds
savefunction to "sox_io" backend, which can save Tensor to a file with the following audio formats;wavmp3flacogg/vorbis** Note The current binary distribution of torchaudio does not containogg/vorbiscodecs. To handle these files, one needs to build torchaudio from the source.