GitLab

Before

Total time: 13.7078

```
   722|    220501|      2.44247|  1.10769e-05| 17.82%|    for i_sample, o0 in enumerate(input_signal_windows.t()):
(call)|         1|  6.36578e-05|  6.36578e-05|  0.00%|# /scratch/moto/pytorch/torch/tensor.py:460 __iter__
(call)|    220500|      1.60566|  7.28191e-06| 11.71%|# /scratch/moto/pytorch/torch/tensor.py:474 <lambda>
   723|    220500|      1.86697|  8.46698e-06| 13.62%|        windowed_output_signal = padded_output_waveform[:, i_sample:(i_sample + n_order)]
   724|    220500|      1.94628|  8.82669e-06| 14.20%|        o0.addmv_(windowed_output_signal, a_coeffs_flipped, alpha=-1)
   725|    220500|         2.46|  1.11565e-05| 17.94%|        o0.div_(a_coeffs[0])
   726|         0|            0|            0|  0.00%|
   727|    220500|      3.37869|  1.53229e-05| 24.64%|        padded_output_waveform[:, i_sample + n_order - 1] = o0
```

After

Total time: 10.9667

```
   722|         1|   9.2268e-05|   9.2268e-05|  0.00%|    input_signal_windows.div_(a_coeffs[0])
   723|         1|  2.14577e-05|  2.14577e-05|  0.00%|    a_coeffs_flipped.div_(a_coeffs[0])
   724|    220501|      2.40216|  1.08941e-05| 21.90%|    for i_sample, o0 in enumerate(input_signal_windows.t()):
(call)|         1|  5.84126e-05|  5.84126e-05|  0.00%|# /scratch/moto/pytorch/torch/tensor.py:460 __iter__
(call)|    220500|      1.59821|   7.2481e-06| 14.57%|# /scratch/moto/pytorch/torch/tensor.py:474 <lambda>
   725|    220500|      1.82273|  8.26633e-06| 16.62%|        windowed_output_signal = padded_output_waveform[:, i_sample:(i_sample + n_order)]
   726|    220500|      1.84074|  8.34802e-06| 16.78%|        o0.addmv_(windowed_output_signal, a_coeffs_flipped, alpha=-1)
   727|    220500|       3.2952|  1.49442e-05| 30.05%|        padded_output_waveform[:, i_sample + n_order - 1] = o0
```

* Add dcshift to functional

* Doc string change and remove inplace clamp

* Minor Fix to dcshit and separate sox test refactoring

* Minor change to limiter_gain type

* adding dcshift to __all__ in functional

* add cmvn

* Update transforms.rst

add cmvn

* Correct the format

* Correct the format

* Correct the format

* add test unit and cmvn change to cmn

* fix bug
Co-authored-by: Vincent QB <vincentqb@users.noreply.github.com>

* Add contrast to functional

* add tests for contrast and update functional.rst

* Minor changes to sox and batch tests for contrast

* Update requirements and travis installation

* Drop support for 3.5 and add test for 3.8

* Call torch.stft directly

* Separate CPU and GPU tests for functions torchscript test

* fix indentation
Co-authored-by: Vincent QB <vincentqb@users.noreply.github.com>

This improves the speed of `lfilter` (and functions that use `lfilter`, such as `biquad`) by 10%.

* Before (23.4369 seconds for `lfilter` call)

Breakdown

```
   720|    220501|       4.4464|   2.0165e-05| 18.97%|    for i_sample, o0 in enumerate(input_signal_windows.t()):
(call)|         1|  7.86781e-05|  7.86781e-05|  0.00%|# /scratch/moto/pytorch/torch/tensor.py:460 __iter__
(call)|    220500|      2.72458|  1.23564e-05| 11.62%|# /scratch/moto/pytorch/torch/tensor.py:474 <lambda>
   721|    220500|      2.80982|   1.2743e-05| 11.99%|        windowed_output_signal = padded_output_waveform[:, i_sample:(i_sample + n_order)]
   722|    220500|      4.92106|  2.23177e-05| 21.00%|        o0.sub_(torch.mv(windowed_output_signal, a_coeffs_flipped))
   723|    220500|      3.72974|  1.69149e-05| 15.91%|        o0.div_(a_coeffs[0])
   724|         0|            0|            0|  0.00%|
   725|    220500|      4.77714|   2.1665e-05| 20.38%|        padded_output_waveform[:, i_sample + n_order - 1] = o0
```

* After (20.8405 seconds for `lfilter` call)

Breakdown

```
   720|    220501|      4.40834|  1.99924e-05| 21.15%|    for i_sample, o0 in enumerate(input_signal_windows.t()):
(call)|         1|  7.31945e-05|  7.31945e-05|  0.00%|# /scratch/moto/pytorch/torch/tensor.py:460 __iter__
(call)|    220500|      2.68595|  1.21812e-05| 12.89%|# /scratch/moto/pytorch/torch/tensor.py:474 <lambda>
   721|    220500|      2.97357|  1.34856e-05| 14.27%|        windowed_output_signal = padded_output_waveform[:, i_sample:(i_sample + n_order)]
   722|    220500|      2.63567|  1.19531e-05| 12.65%|        o0.addmv_(windowed_output_signal, a_coeffs_flipped)
   723|    220500|       3.4228|  1.55229e-05| 16.42%|        o0.div_(a_coeffs[0])
   724|         0|            0|            0|  0.00%|
   725|    220500|      4.68726|  2.12574e-05| 22.49%|        padded_output_waveform[:, i_sample + n_order - 1] = o0
```

* Script

```python
import pprofile

import torch
import torchaudio
import torchaudio.functional as F

def profile_biquad():
    waveform, sr = torchaudio.load('test/assets/100Hz_44100Hz_16bit_05sec.wav', normalization=True)

    prof = pprofile.Profile()
    with prof():
        F.equalizer_biquad(waveform, sr, 3000, 1, 0.707)
    prof.print_stats()

profile_biquad()
```

* See also

https://github.com/pytorch/audio/issues/260#issuecomment-610074110

* add typing to functional

* fix minor things

* fix flake8

* Remove __future__ from torachaudio

* remove double space
Co-authored-by: Vincent QB <vincentqb@users.noreply.github.com>

* add functional DB_to_amplitude

* add test scriptmodule

* add test db_to_amplitude

* add tests

* improve docstrings, move ref for easier use

* Add bandpass to functional.py

* Add bandpass and bandreject to functional

* change name to const_skirt_gain
Co-authored-by: Bhargav Kathivarapu <ka387861@L-156168835.local>
Co-authored-by: Vincent Quenneville-Belair <vincentqb@gmail.com>

* workaround for bartlett_window https://github.com/pytorch/pytorch/issues/32358#issuecomment-576909755

* only change dtype.

* extend batch support

closes #383

* function for batch test.

* set seed.

* adjust tolerance for griffinlim.

* fix power of spectrogram. makes power a float.

closes #389

* commenting out failing test.

* change skip test logic for librosa.
closes #373

* Declare file encoding to support special characters

* fix missing utf_8_encoder error in Travis tests

* Py 2.7 backwards-compat iterator

* ensure integer argument to torch.nn.functional.pad

* cast match.ceil result as integer

* Griffin-Lim Transformation Implementation

* Griffin-Lim Docs

* Remove f-string from backwards compatibility

* iSTFT is now jit-able.

* Comment changes

* Functional Implementation & now jitable

* flake8

* Doc & GPU Fix

* Librosa comparison test

* test directly griffinlim's output. tighter atol.

* matching signature to docstring.
Co-authored-by: Vincent QB <vincentqb@users.noreply.github.com>

* Simplify lfilter functional

* use `torch.clamp` instead of `torch.min(..., torch.max(...))`
* remove unneeded creation of ones tensor for previous method

The current lfilter function uses min and max to essentially do a clamp
function.  I changed the code to use clamp instead.  It is more readable
than the previous version.

FYI, if you want to keep the previous way, you could make a
broadcastable tensor of size 1 instead of creating a tensor the size of
the input.
Signed-off-by: David Pollack <david@da3.net>

* Parallelize waveform windows calculation

I've parallelized the calculation of the waveform windows and also
removed the inefficient calculation within the for-loop.
Signed-off-by: David Pollack <david@da3.net>

* Refactoring and minor readability changes
Signed-off-by: David Pollack <david@da3.net>

* Remove one more creation of a temporary tensor
Signed-off-by: David Pollack <david@da3.net>

* Simpler amplitude_to_DB

* flake8

* Initial commit for SoX logic in VCTK

* change to train whistle file for tests

* apply probability

* fixing errors in docstring.

* move batch to functional.

* nn.Module.

* generalizing spectrogram test.

* adding test to compile functionals.

* add cuda/cpu compilation test.

* adding transform test.

* remove standalone jit file.

* update mel scale.

* remove script decorator.

* apply to augmentations too.

* batching for transforms.

* test for batching.

* update readme.

* STFT transform and function from #285

* merge options in existing functionality.

* remove dimension 2 check. add test.

* using ...

* update spectrogram test.

Add peaking equalizer filter in functional.py and test it in test_functional_filter.py.

* pitch detection validation.
* make torchscriptable.

* Fixed create_fb_matrix filter bank behavior for fmin/fmax
* add better test for f_min close to f_max
* added one more test for f_min > f_max
* adding one more test