Added encoding and bits_per_sample to soundfile's backend save() #1274

prabhat00155 · 2021-02-17T13:23:02Z

Addresses #1258.

mthrok

@prabhat00155 thanks for working on this. The PR looks good overall.

torchaudio/backend/_soundfile_backend.py

test/torchaudio_unittest/backend/soundfile/save_test.py

torchaudio/backend/_soundfile_backend.py

mthrok · 2021-02-19T14:59:34Z

@prabhat00155
Can you merge/rebase on #1285? That should solve the issue with CI.

mthrok

Thanks for the update. I tested the code and it works fine for wav format.

I have not checked flac/vorbis but for sph format there is an issue for subtype generation. Please refer to the comment.

The following is the script I used.

import torch
import torchaudio.backend._soundfile_backend as backend


data = torch.randn(2, 124, dtype=torch.float32)

encs = [
    (None, 8),
    ('PCM_U', None),
    ('PCM_U', 8),
    ('PCM_S', None),
    ('PCM_S', 16),
    ('PCM_S', 32),
    ('PCM_F', None),
    ('PCM_F', 32),
    ('PCM_F', 64),
    ('ULAW', None),
    ('ULAW', 8),
    ('ALAW', None),
    ('ALAW', 8),
]

for enc, bps in encs:
    path = f'tmp3/{enc}_{bps}.wav'
    print(path)
    backend.save(path, data, sample_rate=8000, encoding=enc, bits_per_sample=bps)

dtypes = [
    # torch.uint8,
    torch.int16,
    torch.int32,
    torch.float32,
]
for dtype in dtypes:
    path = f'tmp3/None_None_{dtype}.wav'
    print(path)
    backend.save(path, data.to(dtype), sample_rate=8000)

# bpss = [
#     # 8
#     # 16,
#     # 24,
#     # 32,
#     None,
# ]
# for bps in bpss:
#     path = f'tmp3/{bps}.flac'
#     backend.save(path, data, sample_rate=8000, encoding=None, bits_per_sample=bps)

encs = [
    ('PCM_S', None),
    ('PCM_S', 16),
    ('PCM_S', 32),
    ('ULAW', None),
    ('ULAW', 8),
    ('ALAW', None),
    ('ALAW', 8),
]

for enc, bps in encs:
    path = f'tmp3/{enc}_{bps}.sph'
    print(path)
    backend.save(path, data, sample_rate=8000, encoding=enc, bits_per_sample=bps)

torchaudio/backend/_soundfile_backend.py

test/torchaudio_unittest/backend/soundfile/common.py

mthrok · 2021-02-19T23:58:36Z

Can you generate sph with ALAW encoding? I realized that the resulting files from the script above cannot be opened with soxi.

soxi tmp3/*.sph
soxi FAIL formats: can't open input file `tmp3/ALAW_8.sph': sph: unsupported coding `alaw'
soxi FAIL formats: can't open input file `tmp3/ALAW_None.sph': sph: unsupported coding `alaw'

Input File     : 'tmp3/PCM_S_16.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 16-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 1.52k
Bit Rate       : 785k
Sample Encoding: 16-bit Signed Integer PCM


Input File     : 'tmp3/PCM_S_32.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 32-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 2.02k
Bit Rate       : 1.04M
Sample Encoding: 32-bit Signed Integer PCM


Input File     : 'tmp3/PCM_S_None.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 32-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 2.02k
Bit Rate       : 1.04M
Sample Encoding: 32-bit Signed Integer PCM


Input File     : 'tmp3/ULAW_8.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 14-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 1.27k
Bit Rate       : 657k
Sample Encoding: 8-bit u-law


Input File     : 'tmp3/ULAW_None.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 14-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 1.27k
Bit Rate       : 657k
Sample Encoding: 8-bit u-law

torchaudio/backend/_soundfile_backend.py

prabhat00155 · 2021-02-22T13:51:31Z

Can you generate sph with ALAW encoding? I realized that the resulting files from the script above cannot be opened with soxi.

soxi tmp3/*.sph
soxi FAIL formats: can't open input file `tmp3/ALAW_8.sph': sph: unsupported coding `alaw'
soxi FAIL formats: can't open input file `tmp3/ALAW_None.sph': sph: unsupported coding `alaw'

Input File     : 'tmp3/PCM_S_16.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 16-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 1.52k
Bit Rate       : 785k
Sample Encoding: 16-bit Signed Integer PCM


Input File     : 'tmp3/PCM_S_32.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 32-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 2.02k
Bit Rate       : 1.04M
Sample Encoding: 32-bit Signed Integer PCM


Input File     : 'tmp3/PCM_S_None.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 32-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 2.02k
Bit Rate       : 1.04M
Sample Encoding: 32-bit Signed Integer PCM


Input File     : 'tmp3/ULAW_8.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 14-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 1.27k
Bit Rate       : 657k
Sample Encoding: 8-bit u-law


Input File     : 'tmp3/ULAW_None.sph'
Channels       : 2
Sample Rate    : 8000
Precision      : 14-bit
Duration       : 00:00:00.02 = 124 samples ~ 1.1625 CDDA sectors
File Size      : 1.27k
Bit Rate       : 657k
Sample Encoding: 8-bit u-law

SoX doesn't seem to support ALAW for sphere, check this: https://sourceforge.net/p/sox/code/ci/master/tree/src/sphere.c#l79.

mthrok · 2021-02-22T14:55:06Z

SoX doesn't seem to support ALAW for sphere, check this: https://sourceforge.net/p/sox/code/ci/master/tree/src/sphere.c#l79.

You are right. ffprobe worked. ALAW is working.

ffprobe version 4.2.4-1ubuntu0.1 Copyright (c) 2007-2020 the FFmpeg developers
  built with gcc 9 (Ubuntu 9.3.0-10ubuntu2)
...
Input #0, nistsphere, from 'tmp3/ALAW_8.sph':
  Duration: 00:00:00.02, bitrate: 656 kb/s
    Stream #0:0: Audio: pcm_alaw, 8000 Hz, 2 channels, s16, 128 kb/s

mthrok · 2021-02-23T15:03:37Z

@prabhat00155 Thanks! Can you make (cherry-pick) the same commit and make a PR against release/0.8 branch?

…orch#1274) (cherry picked from commit b8fd5e9)

Cherry picked from commit b8fd5e9 (#1274)

* Add fx graph mode ptq static tuttorial * Add fx graph mode ptq static tuttorial * Remove `_tutorial` from the name so it doesn't build, will add _tutorial after 1.8 Co-authored-by: Brian Johnson <[email protected]>

prabhat00155 added 3 commits February 15, 2021 17:07

Added encoding and bits_per_sample to soundfile's backend save()

8aa45d0

Merge branch 'master' into prabhat00155/modify_save

a385859

Updated unit test

9b1cd44

facebook-github-bot added the CLA Signed label Feb 17, 2021

prabhat00155 added 2 commits February 17, 2021 13:23

Fixed typo

c0440f5

Added newline after docstring list

f60defb

mthrok reviewed Feb 18, 2021

View reviewed changes

prabhat00155 added 2 commits February 18, 2021 12:27

Merge branch 'master' into prabhat00155/modify_save

6ea6074

Addressed review comments

a875fcc

prabhat00155 added 3 commits February 19, 2021 15:08

Merge branch 'master' into prabhat00155/modify_save

3c4b123

Fixed docstring indentation

9bccd7b

Correct flac default

8629e4c

mthrok reviewed Feb 19, 2021

View reviewed changes

torchaudio/backend/_soundfile_backend.py Outdated Show resolved Hide resolved

test/torchaudio_unittest/backend/soundfile/common.py Show resolved Hide resolved

mthrok reviewed Feb 20, 2021

View reviewed changes

torchaudio/backend/_soundfile_backend.py Outdated Show resolved Hide resolved

Added missing _ in subtype for sphere

1b589e9

Revert flac default to PCM_24

354328f

mthrok merged commit b8fd5e9 into pytorch:master Feb 23, 2021

prabhat00155 added a commit to prabhat00155/audio that referenced this pull request Feb 23, 2021

Added encoding and bits_per_sample to soundfile's backend save() (pyt…

dca9ae3

…orch#1274) (cherry picked from commit b8fd5e9)

mthrok mentioned this pull request Feb 23, 2021

cherry pick commits #1274 and #1275 #1302

Closed

prabhat00155 added a commit to prabhat00155/audio that referenced this pull request Feb 23, 2021

Added encoding and bits_per_sample to soundfile's backend save() (pyt…

b708434

…orch#1274) (cherry picked from commit b8fd5e9)

mthrok pushed a commit that referenced this pull request Feb 23, 2021

Added encoding and bits_per_sample to soundfile's backend save() (#1303)

5799599

Cherry picked from commit b8fd5e9 (#1274)

prabhat00155 deleted the prabhat00155/modify_save branch February 23, 2021 18:26

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Added encoding and bits_per_sample to soundfile's backend save() #1274

Added encoding and bits_per_sample to soundfile's backend save() #1274

Uh oh!

prabhat00155 commented Feb 17, 2021

Uh oh!

mthrok left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mthrok commented Feb 19, 2021

Uh oh!

mthrok left a comment

Uh oh!

Uh oh!

Uh oh!

mthrok commented Feb 19, 2021

Uh oh!

Uh oh!

prabhat00155 commented Feb 22, 2021

Uh oh!

mthrok commented Feb 22, 2021

Uh oh!

mthrok commented Feb 23, 2021

Uh oh!

Uh oh!

Added encoding and bits_per_sample to soundfile's backend save() #1274

Added encoding and bits_per_sample to soundfile's backend save() #1274

Uh oh!

Conversation

prabhat00155 commented Feb 17, 2021

Uh oh!

mthrok left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mthrok commented Feb 19, 2021

Uh oh!

mthrok left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

mthrok commented Feb 19, 2021

Uh oh!

Uh oh!

prabhat00155 commented Feb 22, 2021

Uh oh!

mthrok commented Feb 22, 2021

Uh oh!

mthrok commented Feb 23, 2021

Uh oh!

Uh oh!