Files
3d_audio/lib_audio_dsp/test/drc/test_drc_python.py
Steven Dan d8b2974133 init
2025-12-11 09:43:42 +08:00

726 lines
29 KiB
Python

# Copyright 2024-2025 XMOS LIMITED.
# This Software is subject to the terms of the XMOS Public Licence: Version 1.
import numpy as np
import pytest
import soundfile as sf
from pathlib import Path
import os
import audio_dsp.dsp.drc as drc
import audio_dsp.dsp.utils as utils
import audio_dsp.dsp.signal_gen as gen
import audio_dsp.dsp.generic as dspg
from test.test_utils import q_convert_flt
def make_noisy_speech():
hydra_audio_path = os.environ['hydra_audio_PATH']
filepath = Path(hydra_audio_path, 'acoustic_team_test_audio',
'speech', "010_male_female_single-talk_seq.wav")
sig, fs = sf.read(filepath)
amp = utils.db2gain(-30)
noise_sig = gen.pink_noise(fs, len(sig)/fs, amp)
out_sig = sig + noise_sig
out_sig = utils.saturate_float_array(out_sig, dspg.Q_SIG)
return out_sig, fs
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("at", [0.001, 0.01, 0.1, 0.5])
@pytest.mark.parametrize("threshold", [-20, -10, 0])
def test_limiter_peak_attack(fs, at, threshold):
# Attack time test bads on Figure 2 in Guy McNally's "Dynamic Range Control
# of Digital Audio Signals"
if threshold >= dspg.HEADROOM_DB:
pytest.skip("Threshold is above headroom")
return
# Make a constant signal at 6 dB above the threshold, make 2* length of
# attack time to keep the test quick
x = np.ones(int(at*2*fs))
x[:] = utils.db2gain(threshold + 6)
x = utils.saturate_float_array(x, dspg.Q_SIG)
t = np.arange(len(x))/fs
# fixed release time, not that we're going to use it
lt = drc.limiter_peak(fs, 1, threshold, at, 0.3)
y = np.zeros_like(x)
f = np.zeros_like(x)
env = np.zeros_like(x)
# do the processing
for n in range(len(y)):
y[n], f[n], env[n] = lt.process(x[n])
# attack time is defined as how long to get within 2 dB of final value,
# in this case the threshold. Find when we get to this
thresh_passed = np.argmax(utils.db(env) > threshold)
sig_3dB = np.argmax(utils.db(y) < threshold + 2)
measured_at = t[sig_3dB] - t[thresh_passed]
print("target: %.3f, measured: %.3f" % (at, measured_at))
# be somewhere vaugely near the spec, attack time definition is variable!
assert measured_at/at > 0.85
assert measured_at/at < 1.15
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("rt", [0.001, 0.01, 0.1, 0.5])
@pytest.mark.parametrize("threshold", [-20, -10, 0])
def test_limiter_peak_release(fs, rt, threshold):
# Release time test bads on Figure 2 in Guy McNally's "Dynamic Range
# Control of Digital Audio Signals"
if threshold >= dspg.HEADROOM_DB:
pytest.skip("Threshold is above headroom")
return
# Make a step down signal from 6 dB above the threshold to 3 dB below
# threshold, make 2* length of release time plus 0.5 to keep the test quick
x = np.ones(int((0.5+rt*2)*fs))
x[:int(0.5*fs)] = utils.db2gain(threshold + 6)
x[int(0.5*fs):] = utils.db2gain(threshold - 3)
x = utils.saturate_float_array(x, dspg.Q_SIG)
t = np.arange(len(x))/fs
# fixed attack time of 0.01, so should have converged by 0.5s
lt = drc.limiter_peak(fs, 1, threshold, 0.01, rt)
y = np.zeros_like(x)
f = np.zeros_like(x)
env = np.zeros_like(x)
# do the processing
for n in range(len(y)):
y[n], f[n], env[n] = lt.process(x[n])
# find when within 3 dB of target value after dropping below the threshold,
# in this case the original value of 2 dB below the threshold.
sig_3dB = np.argmax(utils.db(y[int(0.5*fs):]) > threshold - 2 - 3)
measured_rt = t[sig_3dB]
print("target: %.3f, measured: %.3f" % (rt, measured_rt))
print(measured_rt/rt)
# be somewhere vaugely near the spec
assert measured_rt/rt > 0.8
assert measured_rt/rt < 1.2
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("threshold", [0, -6, -12])
@pytest.mark.parametrize("ratio", (1, 6, np.inf))
@pytest.mark.parametrize("rt", [0.00000001])
@pytest.mark.parametrize("at", [0.00000001])
def test_comp_ratio(fs, at, rt, ratio, threshold):
# make sure a fast compressor has the same perforance as a limiter
# over a variety of ratios
drcut = drc.compressor_rms(fs, 1, ratio, threshold, at, rt)
signal = gen.log_chirp(fs, (0.1+(rt+at)*2), 1)
signal = utils.saturate_float_array(signal, dspg.Q_SIG)
output_xcore = np.zeros(len(signal))
output_flt = np.zeros(len(signal))
# limiter and compressor have 3 outputs
for n in np.arange(len(signal)):
output_xcore[n], _, _ = drcut.process_xcore(signal[n])
drcut.reset_state()
for n in np.arange(len(signal)):
output_flt[n], _, _ = drcut.process(signal[n])
# lazy limiter
ref_signal = np.copy(signal)
over_thresh = utils.db(ref_signal) > threshold
ref_signal[over_thresh] *= utils.db2gain((1 - 1/ratio)*(threshold - utils.db(ref_signal[over_thresh])))
np.testing.assert_allclose(ref_signal, output_flt, atol=3e-16, rtol=0)
# threshold limited by 24b resolution in float32 mantissa
np.testing.assert_allclose(output_flt, output_xcore, atol=2**-23, rtol=0)
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("at", [0.001, 0.01, 0.1, 0.5])
@pytest.mark.parametrize("threshold", [-20, -10, 0])
def comp_vs_limiter(fs, at, threshold):
# check infinite ratio compressor is a limiter
# Make a constant signal at 6dB above the threshold, make 2* length of
# attack time to keep the test quick
x = np.ones(int(at*2*fs))
x[:] = utils.db2gain(threshold + 6)
x = utils.saturate_float_array(x, dspg.Q_SIG)
t = np.arange(len(x))/fs
rt = 0.3
comp_type = "rms"
comp_handle = getattr(drc, "compressor_%s" % comp_type)
lim_handle = getattr(drc, "limiter_%s" % comp_type)
comp_thing = comp_handle(fs, 1, threshold, np.inf, at, rt)
lim_thing = lim_handle(fs, 1, threshold, at, rt)
y_p = np.zeros_like(x)
f_p = np.zeros_like(x)
env_p = np.zeros_like(x)
# do the processing
for n in range(len(y_p)):
y_p[n], f_p[n], env_p[n] = comp_thing.process(x[n])
y_r = np.zeros_like(x)
f_r = np.zeros_like(x)
env_r = np.zeros_like(x)
# do the processing
for n in range(len(y_r)):
y_r[n], f_r[n], env_r[n] = lim_thing.process(x[n])
# limiter and infinite ratio compressor should be the same
np.testing.assert_allclose(utils.db(y_p),
utils.db(y_r),
atol=0.002)
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("at", [0.001, 0.01, 0.1, 0.5])
@pytest.mark.parametrize("threshold", [-20, -10, 0])
def test_peak_vs_rms(fs, at, threshold):
# check peak and rms converge to same value
# Make a constant signal at 6dB above the threshold, make 2* length of
# attack time to keep the test quick
x = np.ones(int(at*10*fs))
x[:] = utils.db2gain(threshold + 6)
x = utils.saturate_float_array(x, dspg.Q_SIG)
t = np.arange(len(x))/fs
rt = 0.3
comp_type = "limiter"
peak_handle = getattr(drc, "%s_peak" % comp_type)
rms_handle = getattr(drc, "%s_rms" % comp_type)
peak_thing = peak_handle(fs, 1, threshold, at, rt)
rms_thing = rms_handle(fs, 1, threshold, at, rt)
y_p = np.zeros_like(x)
f_p = np.zeros_like(x)
env_p = np.zeros_like(x)
# do the processing
for n in range(len(y_p)):
y_p[n], f_p[n], env_p[n] = peak_thing.process(x[n])
y_r = np.zeros_like(x)
f_r = np.zeros_like(x)
env_r = np.zeros_like(x)
# do the processing
for n in range(len(y_r)):
y_r[n], f_r[n], env_r[n] = rms_thing.process(x[n])
# rms and peak limiter should converge to the same value
np.testing.assert_allclose(utils.db(y_p[int(fs*at*5):]),
utils.db(y_r[int(fs*at*5):]),
atol=0.0022)
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("at", [0.01])
@pytest.mark.parametrize("threshold", [-10])
def test_sidechain_mono_vs_comp(fs, at, threshold):
# test a sidechain compressor is the same as a normal compressor
# when the sidechain signal is the same as the input signal
ratio = 5
x = gen.sin(fs, 1, 1, 1)
x = utils.saturate_float_array(x, dspg.Q_SIG)
t = np.arange(len(x))/fs
rt = 0.5
comp_type = "compressor_rms"
reg_handle = getattr(drc, "%s" % comp_type)
side_handle = getattr(drc, "%s_sidechain_mono" % comp_type)
reg_thing = reg_handle(fs, 1, ratio, threshold, at, rt)
side_thing = side_handle(fs, ratio, threshold, at, rt)
y_p = np.zeros_like(x)
f_p = np.zeros_like(x)
env_p = np.zeros_like(x)
# do the processing
for n in range(len(y_p)):
y_p[n], f_p[n], env_p[n] = reg_thing.process(x[n])
y_r = np.zeros_like(x)
f_r = np.zeros_like(x)
env_r = np.zeros_like(x)
# do the processing
for n in range(len(y_r)):
y_r[n], f_r[n], env_r[n] = side_thing.process(x[n], x[n])
# rms and peak limiter should converge to the same value
np.testing.assert_array_equal(y_p, y_r)
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("at", [0.01])
@pytest.mark.parametrize("threshold", [-10])
def test_sidechain_stereo(fs, at, threshold):
# check peak and rms converge to same value
# Make a constant signal at 6dB above the threshold, make 2* length of
# attack time to keep the test quick
x = gen.sin(fs, 1, 1, 1)
x = utils.saturate_float_array(x, dspg.Q_SIG)
x = np.stack([x, x], axis=0)
t = np.arange(len(x))/fs
rt = 0.3
comp_type = "compressor_rms"
reg_handle = getattr(drc, "%s_stereo" % comp_type)
side_handle = getattr(drc, "%s_sidechain_stereo" % comp_type)
reg_thing = reg_handle(fs, 1, threshold, at, rt)
side_thing = side_handle(fs, 1, threshold, at, rt)
y_p = np.zeros_like(x)
f_p = np.zeros_like(x)
env_p = np.zeros_like(x)
# do the processing
for n in range(len(y_p)):
y_p[:, n], f_p[:, n], env_p[:, n] = reg_thing.process_channels(x[:, n])
y_r = np.zeros_like(x)
f_r = np.zeros_like(x)
env_r = np.zeros_like(x)
# do the processing
for n in range(len(y_r)):
y_r[:, n], f_r[:, n], env_r[:, n] = side_thing.process_channels(x[:, n], x[:, n])
# rms and peak limiter should converge to the same value
np.testing.assert_array_equal(y_p, y_r)
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("component_mono, component_stereo, threshold, ratio", [("limiter_peak", "limiter_peak_stereo", -20, None),
("limiter_peak", "limiter_peak_stereo", -6, None),
("compressor_rms", "compressor_rms_stereo", 0, 6),
("compressor_rms", "compressor_rms_stereo", 0, 2),
("compressor_rms_sidechain_mono", "compressor_rms_sidechain_stereo", 0, 6),
("compressor_rms_sidechain_mono", "compressor_rms_sidechain_stereo", 0, 2)])
@pytest.mark.parametrize("rt", [0.2])
@pytest.mark.parametrize("at", [0.001])
def test_mono_vs_stereo(fs, component_mono, component_stereo, at, rt, threshold, ratio):
# test the mono and stereo components have the same perforamnce when
# fed a dual mono signal
signal = []
length = 0.1 + (rt + at) * 2
f = 997
signal.append(gen.sin(fs, length, f, 1))
signal.append(gen.sin(fs, length, f, 1))
signal = np.stack(signal, axis=0)
signal = utils.saturate_float_array(signal, dspg.Q_SIG)
if "sidechain" in component_stereo:
sidechain_signal = np.zeros_like(signal)
sidechain_signal[:, len(signal)//2:] = 1
sidechain_signal = utils.saturate_float_array(sidechain_signal, dspg.Q_SIG)
stereo_component_handle = getattr(drc, component_stereo)
mono_component_handle = getattr(drc, component_mono)
if ratio is not None:
drc_s = stereo_component_handle(fs, ratio, threshold, at, rt)
if "sidechain" in component_mono:
drc_m = mono_component_handle(fs, ratio, threshold, at, rt)
else:
drc_m = mono_component_handle(fs, 1, ratio, threshold, at, rt)
else:
drc_s = stereo_component_handle(fs, threshold, at, rt)
drc_m = mono_component_handle(fs, 1, threshold, at, rt)
output_xcore_s = np.zeros(signal.shape)
output_flt_s = np.zeros(signal.shape)
if "sidechain" in component_stereo:
for n in np.arange(signal.shape[1]):
output_xcore_s[:, n], _, _ = drc_s.process_channels_xcore(signal[:, n], sidechain_signal[:, n])
drc_s.reset_state()
for n in np.arange(signal.shape[1]):
output_flt_s[:, n], _, _ = drc_s.process_channels(signal[:, n], sidechain_signal[:, n])
else:
for n in np.arange(signal.shape[1]):
output_xcore_s[:, n], _, _ = drc_s.process_channels_xcore(signal[:, n])
drc_s.reset_state()
for n in np.arange(signal.shape[1]):
output_flt_s[:, n], _, _ = drc_s.process_channels(signal[:, n])
drc_s.reset_state()
output_xcore_m = np.zeros(signal.shape)
output_flt_m = np.zeros(signal.shape)
# write mono signal to both output channels, makes comparison to stereo easier
if "sidechain" in component_mono:
for n in np.arange(signal.shape[1]):
output_xcore_m[:, n], _, _ = drc_m.process_xcore(signal[0, n], sidechain_signal[0, n])
drc_m.reset_state()
for n in np.arange(signal.shape[1]):
output_flt_m[:, n], _, _ = drc_m.process(signal[0, n], sidechain_signal[0, n])
else:
for n in np.arange(signal.shape[1]):
output_xcore_m[:, n], _, _ = drc_m.process_xcore(signal[0, n])
drc_m.reset_state()
for n in np.arange(signal.shape[1]):
output_flt_m[:, n], _, _ = drc_m.process(signal[0, n])
# check stereo channels are the same
np.testing.assert_array_equal(output_flt_s[0], output_flt_s[1])
np.testing.assert_array_equal(output_xcore_s[0], output_xcore_s[1])
# check stereo equals mono
#TODO this should be equal
np.testing.assert_allclose(output_flt_s, output_flt_m, atol=2**-32)
#TODO this should be array_equal
np.testing.assert_allclose(output_xcore_s, output_xcore_m, atol=2**-31)
@pytest.mark.parametrize("component, threshold, ratio", [("noise_gate", -30, None),
("noise_suppressor_expander", -30, 3)])
def test_noise_gate(component, threshold, ratio):
# test the noise gate performance on noisy speech
signal, fs = make_noisy_speech()
test_len = int(6*fs)
signal = signal[:test_len]
signal = utils.saturate_float_array(signal, dspg.Q_SIG)
if ratio:
drcut = drc.noise_suppressor_expander(fs, 1, ratio, threshold, 0.005, 0.2)
else:
drcut = drc.noise_gate(fs, 1, threshold, 0.005, 0.2)
output_xcore = np.zeros(len(signal))
output_flt = np.zeros(len(signal))
gain_xcore = np.zeros(len(signal))
gain_flt = np.zeros(len(signal))
env_xcore = np.zeros(len(signal))
env_flt = np.zeros(len(signal))
# noise gate has 3 outputs
for n in np.arange(len(signal)):
output_xcore[n], gain_xcore[n], env_xcore[n] = drcut.process_xcore(signal[n])
drcut.reset_state()
for n in np.arange(len(signal)):
output_flt[n], gain_flt[n], env_flt[n] = drcut.process(signal[n])
sf.write("%s_test_in.wav" % component, signal, fs)
sf.write("%s_test_out.wav" % component, output_flt, fs)
# small signals are always going to be ropey due to quantizing, so just check average error of top half
top_half = utils.db(output_flt) > -50
if np.any(top_half):
error_flt = np.abs(utils.db(output_xcore[top_half])-utils.db(output_flt[top_half]))
mean_error_flt = utils.db(np.nanmean(utils.db2gain(error_flt)))
assert mean_error_flt < 0.055
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("component, threshold, ratio", [("limiter_peak", 0, None),
("limiter_rms", 0, None),
("compressor_rms", 0, 6),
("compressor_rms", 0, 2),
("compressor_rms_softknee", 6, 6),
("compressor_rms_softknee", 6, 2),
("noise_gate", -1000, None),
("noise_suppressor_expander", -1000, 5),
("hard_limiter_peak", 0, None),
("clipper", 0, None)])
@pytest.mark.parametrize("rt, at", [[0.001, 0.2], [0.1, 0.5]])
def test_drc_component_bypass(fs, component, at, rt, threshold, ratio):
# test that a drc component is bit exact when the signal is below
# the threshold (or above in the case of a noise gate).
component_handle = getattr(drc, component)
if threshold is not None:
if ratio is not None:
drcut = component_handle(fs, 1, ratio, threshold, at, rt)
elif "clipper" in component:
drcut = component_handle(fs, 1, threshold)
else:
drcut = component_handle(fs, 1, threshold, at, rt)
else:
drcut = component_handle(fs, 1, at, rt)
if "softknee" in component:
# soft knee acts below threshold, so needs to be smaller for
# bypass test
signal = gen.log_chirp(fs, (0.1+(rt+at)*2), 0.5)
else:
signal = gen.log_chirp(fs, (0.1+(rt+at)*2), 1)
signal = q_convert_flt(signal, 23, dspg.Q_SIG)
signal = utils.saturate_float_array(signal, dspg.Q_SIG)
output_xcore = np.zeros(len(signal))
output_flt = np.zeros(len(signal))
if "envelope" in component or "clipper" in component:
# envelope and clipper have 1 output
for n in np.arange(len(signal)):
output_xcore[n] = drcut.process_xcore(signal[n])
if "clipper" not in component:
drcut.reset_state()
for n in np.arange(len(signal)):
output_flt[n] = drcut.process(signal[n])
else:
# limiter and compressor have 3 outputs
for n in np.arange(len(signal)):
output_xcore[n], _, _ = drcut.process_xcore(signal[n])
drcut.reset_state()
for n in np.arange(len(signal)):
output_flt[n], _, _ = drcut.process(signal[n])
np.testing.assert_array_equal(signal, output_flt)
np.testing.assert_allclose(signal, output_xcore, atol=2**-32)
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("component, threshold, ratio", [("limiter_peak", -20, None),
("limiter_peak", 6, None),
("limiter_rms", -20, None),
("limiter_rms", 6, None),
("envelope_detector_peak", None, None),
("envelope_detector_rms", None, None),
("compressor_rms", -20, 6),
("compressor_rms", -20, 2),
("compressor_rms", 6, 6),
("compressor_rms", 6, 2),
("compressor_rms_softknee", -20, 6),
("compressor_rms_softknee", -20, 2),
("compressor_rms_softknee", 6, 6),
("compressor_rms_softknee", 6, 2),
("noise_gate", -20, None),
("noise_suppressor_expander", -20, 5),
("hard_limiter_peak", -20, None),
("hard_limiter_peak", 6, None),
("clipper", -20, None),
("clipper", 6, None)])
@pytest.mark.parametrize("rt, at", [[0.001, 0.2], [0.1, 0.5], [0.001, 0.5], [0.1, 0.2]])
def test_drc_component(fs, component, at, rt, threshold, ratio):
# test the process_ functions of the drc components
component_handle = getattr(drc, component)
if threshold is not None:
if ratio is not None:
drcut = component_handle(fs, 1, ratio, threshold, at, rt)
elif "clipper" in component:
drcut = component_handle(fs, 1, threshold)
else:
drcut = component_handle(fs, 1, threshold, at, rt)
else:
drcut = component_handle(fs, 1, at, rt)
signal = gen.log_chirp(fs, (0.1+(rt+at)*2), 1)
len_sig = len(signal)
if threshold is not None:
# If we are a limiter or compressor, have first half of signal above
# the threshold and second half below
signal[:len_sig//2] *= utils.db2gain(threshold + 6)
signal[len_sig//2:] *= utils.db2gain(threshold - 3)
else:
# if we are an envelope detector, amplitude modulate with a sin to give
# something to follow
t = np.arange(len(signal))/fs
signal *= np.sin(t*2*np.pi*0.5)
signal = q_convert_flt(signal, 23, dspg.Q_SIG)
signal = utils.saturate_float_array(signal, dspg.Q_SIG)
output_xcore = np.zeros(len(signal))
output_flt = np.zeros(len(signal))
gain_xcore = np.zeros(len(signal))
gain_flt = np.zeros(len(signal))
env_xcore = np.zeros(len(signal))
env_flt = np.zeros(len(signal))
if "envelope" in component or "clipper" in component:
# envelope and clipper have 1 output
for n in np.arange(len(signal)):
output_xcore[n] = drcut.process_xcore(signal[n])
if "clipper" not in component:
drcut.reset_state()
for n in np.arange(len(signal)):
output_flt[n] = drcut.process(signal[n])
else:
# limiter and compressor have 3 outputs
for n in np.arange(len(signal)):
output_xcore[n], gain_xcore[n], env_xcore[n] = drcut.process_xcore(signal[n])
drcut.reset_state()
for n in np.arange(len(signal)):
output_flt[n], gain_flt[n], env_flt[n] = drcut.process(signal[n])
# small signals are always going to be ropey due to quantizing, so just check average error of top half
top_half = utils.db(output_flt) > -50
if np.any(top_half):
error_flt = np.abs(utils.db(output_xcore[top_half])-utils.db(output_flt[top_half]))
mean_error_flt = utils.db(np.nanmean(utils.db2gain(error_flt)))
if "softknee" in component:
# different implementation of knee adds more error
if drcut.Q_sig > 27:
pytest.skip("Soft knee breaks above Q4.27")
else:
assert mean_error_flt < 0.08
else:
assert mean_error_flt < 0.055
if "hard" in component or "clip" in component:
# roudning error can occur in threshold
assert np.all(output_xcore <= utils.db2gain(threshold) + 2**-(drcut.Q_sig + 1))
assert np.all(output_flt <= utils.db2gain(threshold))
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("component, threshold, ratio", [("limiter_peak", -20, None),
("limiter_peak", 6, None),
("limiter_rms", -20, None),
("limiter_rms", 6, None),
("envelope_detector_peak", None, None),
("envelope_detector_rms", None, None),
("compressor_rms", -20, 6),
("compressor_rms", -20, 2),
("compressor_rms", 6, 6),
("compressor_rms", 6, 2),
("compressor_rms_softknee", -20, 6),
("compressor_rms_softknee", -20, 2),
("compressor_rms_softknee", 6, 6),
("compressor_rms_softknee", 6, 2),
("noise_gate", -20, None),
("noise_suppressor_expander", -20, 5),
("hard_limiter_peak", -20, None),
("hard_limiter_peak", 6, None),
("clipper", -20, None),
("clipper", 6, None)])
@pytest.mark.parametrize("rt, at", [[0.001, 0.2], [0.1, 0.5]])
@pytest.mark.parametrize("n_chans", [1, 2, 4])
@pytest.mark.parametrize("q_format", [27, 31])
def test_drc_component_frames(fs, component, at, rt, threshold, ratio, n_chans, q_format):
# test the process_frame functions of the drc components
if q_format == 31 and rt != 0.2 and at != 0.001 and n_chans != 1:
pytest.skip("Don't run all tests at Q0.31")
component_handle = getattr(drc, component)
if threshold is not None:
if ratio is not None:
drcut = component_handle(fs, n_chans, ratio, threshold, at, rt, Q_sig=q_format)
elif "clipper" in component:
drcut = component_handle(fs, n_chans, threshold, Q_sig=q_format)
else:
drcut = component_handle(fs, n_chans, threshold, at, rt, Q_sig=q_format)
else:
drcut = component_handle(fs, n_chans, at, rt)
signal = gen.log_chirp(fs, (0.1+(rt+at)*2), 1)
len_sig = len(signal)
if threshold is not None:
signal[:len_sig//2] *= utils.db2gain(threshold + 6)
signal[len_sig//2:] *= utils.db2gain(threshold - 3)
else:
t = np.arange(len(signal))/fs
signal *= np.sin(t*2*np.pi*0.5)
signal = utils.saturate_float_array(signal, q_format)
signal = np.tile(signal, [n_chans, 1])
frame_size = 1
signal_frames = utils.frame_signal(signal, frame_size, 1)
output_int = np.zeros_like(signal)
output_flt = np.zeros_like(signal)
for n in range(len(signal_frames)):
output_int[:, n*frame_size:(n+1)*frame_size] = drcut.process_frame_xcore(signal_frames[n])
if "clipper" not in component:
drcut.reset_state()
for n in range(len(signal_frames)):
output_flt[:, n*frame_size:(n+1)*frame_size] = drcut.process_frame(signal_frames[n])
assert np.all(output_int[0, :] == output_int)
assert np.all(output_flt[0, :] == output_flt)
@pytest.mark.parametrize("fs", [48000])
@pytest.mark.parametrize("component, threshold, ratio", [("limiter_peak_stereo", -20, None),
("limiter_peak_stereo", -6, None),
("compressor_rms_stereo", 0, 6),
("compressor_rms_stereo", 0, 2)])
@pytest.mark.parametrize("rt, at", [[0.001, 0.2], [0.1, 0.5]])
def test_stereo_components(fs, component, at, rt, threshold, ratio):
# test the process_channels functions of the stereo drc components
component_handle = getattr(drc, component)
if ratio is not None:
drcut = component_handle(fs, ratio, threshold, at, rt)
else:
drcut = component_handle(fs, threshold, at, rt)
signal = []
length = 0.1 + (rt + at) * 2
f = 997
signal.append(gen.sin(fs, length, f, 1))
signal.append(gen.sin(fs, length, f, 0.5))
signal = np.stack(signal, axis=0)
signal = utils.saturate_float_array(signal, dspg.Q_SIG)
output_xcore = np.zeros(signal.shape)
output_flt = np.zeros(signal.shape)
for n in np.arange(signal.shape[1]):
output_xcore[:, n], _, _ = drcut.process_channels_xcore(signal[:, n])
drcut.reset_state()
for n in np.arange(signal.shape[1]):
output_flt[:, n], _, _ = drcut.process_channels(signal[:, n])
error_flt = np.abs(utils.db(output_xcore)-utils.db(output_flt))
mean_error_flt = utils.db(np.nanmean(utils.db2gain(error_flt)))
assert mean_error_flt < 0.055
# TODO more RMS limiter tests
# TODO hard limiter test
# TODO envelope detector tests
# TODO compressor tests
if __name__ == "__main__":
# test_drc_component_frames(48000, "compressor_rms", 0.1, 0.5, 6, 6, 1, 31)
# test_limiter_peak_attack(48000, 0.001, 0)
# comp_vs_limiter(48000, 0.001, 0)
# test_comp_ratio(48000, 0.00000001, 0.00000001, 2, 0)
test_mono_vs_stereo(48000, "compressor_rms_sidechain_mono", "compressor_rms_sidechain_stereo", 0.001, 0.01, 0, 6)
# test_sidechain_mono_vs_comp(16000, 0.05, -40)
# test_noise_gate("noise_gate", -30, None)
# test_drc_component_bypass(48000, "compressor_rms", 0.01, 0.2, 0, 6)