Skip to content

Add peaking equalizer filter #315

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Nov 1, 2019
Merged

Add peaking equalizer filter #315

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
368e943
Add peaking equalizer filter in functional.py and test it in test_fun…
xinyang0 Oct 25, 2019
cc343e0
Remove whitespace in black line
xinyang0 Oct 25, 2019
96c4892
Revome A=1 special case. Add a default value for Q. Decrease center_f…
xinyang0 Nov 1, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 20 additions & 0 deletions test/test_functional_filtering.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -136,6 +136,26 @@ def test_highpass(self):
# TBD - this fails at the 1e-4 level, debug why
assert torch.allclose(sox_output_waveform, output_waveform, atol=1e-3)

def test_equalizer(self):
"""
Test biquad peaking equalizer filter, compare to SoX implementation
"""

CENTER_FREQ = 300
Q = 0.707
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a chance that these values are special and miss something important?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good question! The specgram of the white noise signal is actually in the range of [0, 578], so 1000 is a bit large for this signal. Decrease it to 300.

GAIN = 1

noise_filepath = os.path.join(self.test_dirpath, "assets", "whitenoise.mp3")
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@vincentqb instead of calling into sox while testing, should we write out reference files instead? especially since the asset is fixed.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, I agree with you.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Although this could be done in a separate PR for all the other tests too.

E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(noise_filepath)
E.append_effect_to_chain("equalizer", [CENTER_FREQ, Q, GAIN])
sox_output_waveform, sr = E.sox_build_flow_effects()

waveform, sample_rate = torchaudio.load(noise_filepath, normalization=True)
output_waveform = F.equalizer_biquad(waveform, sample_rate, CENTER_FREQ, GAIN, Q)

assert torch.allclose(sox_output_waveform, output_waveform, atol=1e-4)

def test_perf_biquad_filtering(self):

fn_sine = os.path.join(self.test_dirpath, "assets", "whitenoise.mp3")
Expand Down
28 changes: 28 additions & 0 deletions torchaudio/functional.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@
"lfilter",
"lowpass_biquad",
"highpass_biquad",
"equalizer_biquad",
"biquad",
'mask_along_axis',
'mask_along_axis_iid'
Expand Down Expand Up @@ -683,6 +684,33 @@ def lowpass_biquad(waveform, sample_rate, cutoff_freq, Q=0.707):
return biquad(waveform, b0, b1, b2, a0, a1, a2)


def equalizer_biquad(waveform, sample_rate, center_freq, gain, Q=0.707):
# type: (Tensor, int, float, float, float) -> Tensor
r"""Designs biquad peaking equalizer filter and performs filtering. Similar to SoX implementation.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure its necessary to mention that this is similar to Sox. Its a well-defined operation that stands by itself.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agree. It's mentioned in the other transforms, so we'll have to remove it elsewhere too.


Args:
waveform (torch.Tensor): audio waveform of dimension of `(channel, time)`
sample_rate (int): sampling rate of the waveform, e.g. 44100 (Hz)
center_freq (float): filter’s central frequency
gain (float): desired gain at the boost (or attenuation) in dB
q_factor (float): https://en.wikipedia.org/wiki/Q_factor

Returns:
output_waveform (torch.Tensor): Dimension of `(channel, time)`
"""
w0 = 2 * math.pi * center_freq / sample_rate
A = math.exp(gain / 40.0 * math.log(10))
alpha = math.sin(w0) / 2 / Q

b0 = 1 + alpha * A
b1 = -2 * math.cos(w0)
b2 = 1 - alpha * A
a0 = 1 + alpha / A
a1 = -2 * math.cos(w0)
a2 = 1 - alpha / A
return biquad(waveform, b0, b1, b2, a0, a1, a2)


@torch.jit.script
def mask_along_axis_iid(specgrams, mask_param, mask_value, axis):
# type: (Tensor, int, float, int) -> Tensor
Expand Down