Add flanger to functional

[bhargavkathivarapu]

Hi ,
Regarding the issue ( Reducing dependency in Sox - #260 )
Implemented flanger function in functional.py

• function.rst - updated with new function

• test_sox_compatibility.py - Added four tests for flanger ( Tested if all variations are working fine . we can remove some tests as it adds more time to test run time )

• test_batch_consistency.py - added test for flanger

• test_torchscript_consistency.py - added test for flanger

All CPU test passed.

@vincentqb or @mthrok could you please review these changes

[codecov]

Codecov Report

Merging #629 into master will increase coverage by 0.30%.
The diff coverage is 98.21%.

@@            Coverage Diff             @@
##           master     #629      +/-   ##
==========================================
+ Coverage   89.00%   89.31%   +0.30%     
==========================================
  Files          21       21              
  Lines        2255     2311      +56     
==========================================
+ Hits         2007     2064      +57     
+ Misses        248      247       -1     

Impacted Files	Coverage Δ
torchaudio/functional.py	95.94% <98.21%> (+0.43%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 00d3820...7440564. Read the comment docs.

[mthrok]

Correct me if I am wrong, it seems that the same operation with different parameters is applied to each channel individually. If operations among channels do not interweave, can we vectorize the operation?

Looking at the test result, the test of this function takes the longest.
https://app.circleci.com/pipelines/github/pytorch/audio/1768/workflows/4a2bae91-b123-4fcd-af13-a0a24c929c1e/jobs/44621

[bhargavkathivarapu]

@mthrok , I have vectorized the channels loop , But facing issue in torch script test , I am getting below error

RuntimeError: 
E           Arguments for call are not valid.
E           The following variants are available:
E             
E             aten::frac(Tensor self) -> (Tensor):
E             Expected a value of type 'Tensor' for argument 'self' but instead found type 'float'.
E             
E             aten::frac.out(Tensor self, *, Tensor(a!) out) -> (Tensor(a!)):
E             Expected a value of type 'Tensor' for argument 'self' but instead found type 'float'.
E           
E           The original call is:
E             File "/Users/bhargav/audio/torchaudio/functional.py", line 1542
E                   cur_channel_phase = (torch.arange(0, n_channels) * lfo_length * channel_phase + .5).to(torch.int64)
E                   delay = lfo[(lfo_pos + cur_channel_phase) % lfo_length]
E                   frac_delay = torch.frac(delay)
E                                ~~~~~~~~~~ <--- HERE
E                   delay = torch.floor(delay)
E           'flanger' is being compiled since it was called from 'func'
E             File "/Users/bhargav/audio/test/test_torchscript_consistency.py", line 542
E                       phase = 60.
E                       sample_rate = 44100
E                       return F.flanger(tensor, sample_rate, delay, depth, regen, width, speed,
E                              ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~...  <--- HERE
E                                        phase, sinusoidal=True, linear_interpolation=True)

The delay variable is a tensor, But the torch script is saying it got float instead of tensor

[mthrok]

delay is used in function signature and annotated as float, so JIT compiler considers it float but you are also re-defining variable delay as Tensor.

Change the variable name to something like delay_tensor = lfo[(lfo_pos + cur_channel_phase) % lfo_length], and it should work.

[bhargavkathivarapu]

@mthrok , Fixed it , Torch script test passed

[mthrok]

@bhargavkathivarapu
Thanks. Can you rebase? In #644, I moved the test file.

Also, in Torchscript test and batch consistency test, instead of loading sample value from a file, can you generate input value, with fixed random seed?

[bhargavkathivarapu]

@mthrok , I have messed up rebasing , I am closing this pull request . Made a new pull request with suggested changes( #651 ) , please check that .