camenduru's picture
thanks to RVC-Project ❤
ef99749
raw
history blame
No virus
21.7 kB
import numpy as np
from joblib import Parallel, delayed
import librosa
from scipy.signal import filtfilt
import scipy
import tempfile
from tqdm.auto import tqdm
def sigmoid(x, shift, mult):
"""
Using this sigmoid to discourage one network overpowering the other
"""
return 1 / (1 + np.exp(-(x + shift) * mult))
def get_time_smoothed_representation(
spectral, samplerate, hop_length, time_constant_s=0.001
):
t_frames = time_constant_s * samplerate / float(hop_length)
# By default, this solves the equation for b:
# b**2 + (1 - b) / t_frames - 2 = 0
# which approximates the full-width half-max of the
# squared frequency response of the IIR low-pass filt
b = (np.sqrt(1 + 4 * t_frames ** 2) - 1) / (2 * t_frames ** 2)
return filtfilt([b], [1, b - 1], spectral, axis=-1, padtype=None)
def _smoothing_filter(n_grad_freq, n_grad_time):
"""Generates a filter to smooth the mask for the spectrogram
Arguments:
n_grad_freq {[type]} -- [how many frequency channels to smooth over with the mask.]
n_grad_time {[type]} -- [how many time channels to smooth over with the mask.]
"""
smoothing_filter = np.outer(
np.concatenate(
[
np.linspace(0, 1, n_grad_freq + 1, endpoint=False),
np.linspace(1, 0, n_grad_freq + 2),
]
)[1:-1],
np.concatenate(
[
np.linspace(0, 1, n_grad_time + 1, endpoint=False),
np.linspace(1, 0, n_grad_time + 2),
]
)[1:-1],
)
smoothing_filter = smoothing_filter / np.sum(smoothing_filter)
return smoothing_filter
class SpectralGate:
def __init__(
self,
y,
sr,
prop_decrease,
chunk_size,
padding,
n_fft,
win_length,
hop_length,
time_constant_s,
freq_mask_smooth_hz,
time_mask_smooth_ms,
tmp_folder,
use_tqdm,
n_jobs
):
self.sr = sr
# if this is a 1D single channel recording
self.flat = False
y = np.array(y)
# reshape data to (#channels, #frames)
if len(y.shape) == 1:
self.y = np.expand_dims(y, 0)
self.flat = True
elif len(y.shape) > 2:
raise ValueError("Waveform must be in shape (# frames, # channels)")
else:
self.y = y
self._dtype = y.dtype
# get the number of channels and frames in data
self.n_channels, self.n_frames = self.y.shape
self._chunk_size = chunk_size
self.padding = padding
self.n_jobs = n_jobs
self.use_tqdm = use_tqdm
# where to create a temp file for parallel
# writing
self._tmp_folder = tmp_folder
### Parameters for spectral gating
self._n_fft = n_fft
# set window and hop length for stft
if win_length is None:
self._win_length = self._n_fft
else:
self._win_length = win_length
if hop_length is None:
self._hop_length = self._win_length // 4
else:
self._hop_length = hop_length
self._time_constant_s = time_constant_s
self._prop_decrease = prop_decrease
if (freq_mask_smooth_hz is None) & (time_mask_smooth_ms is None):
self.smooth_mask = False
else:
self._generate_mask_smoothing_filter(freq_mask_smooth_hz, time_mask_smooth_ms)
def _generate_mask_smoothing_filter(self, freq_mask_smooth_hz, time_mask_smooth_ms):
if freq_mask_smooth_hz is None:
n_grad_freq = 1
else:
# filter to smooth the mask
n_grad_freq = int(freq_mask_smooth_hz / (self.sr / (self._n_fft / 2)))
if n_grad_freq < 1:
raise ValueError(
"freq_mask_smooth_hz needs to be at least {}Hz".format(
int((self.sr / (self._n_fft / 2)))
)
)
if time_mask_smooth_ms is None:
n_grad_time = 1
else:
n_grad_time = int(time_mask_smooth_ms / ((self._hop_length / self.sr) * 1000))
if n_grad_time < 1:
raise ValueError(
"time_mask_smooth_ms needs to be at least {}ms".format(
int((self._hop_length / self.sr) * 1000)
)
)
if (n_grad_time == 1) & (n_grad_freq == 1):
self.smooth_mask = False
else:
self.smooth_mask = True
self._smoothing_filter = _smoothing_filter(n_grad_freq, n_grad_time)
def _read_chunk(self, i1, i2):
"""read chunk and pad with zerros"""
if i1 < 0:
i1b = 0
else:
i1b = i1
if i2 > self.n_frames:
i2b = self.n_frames
else:
i2b = i2
chunk = np.zeros((self.n_channels, i2 - i1))
chunk[:, i1b - i1 : i2b - i1] = self.y[:, i1b:i2b]
return chunk
def filter_chunk(self, start_frame, end_frame):
"""Pad and perform filtering"""
i1 = start_frame - self.padding
i2 = end_frame + self.padding
padded_chunk = self._read_chunk(i1, i2)
filtered_padded_chunk = self._do_filter(padded_chunk)
return filtered_padded_chunk[:, start_frame - i1 : end_frame - i1]
def _get_filtered_chunk(self, ind):
"""Grabs a single chunk"""
start0 = ind * self._chunk_size
end0 = (ind + 1) * self._chunk_size
return self.filter_chunk(start_frame=start0, end_frame=end0)
def _do_filter(self, chunk):
"""Do the actual filtering"""
raise NotImplementedError
return chunk_filtered
def _iterate_chunk(self, filtered_chunk, pos, end0, start0, ich):
filtered_chunk0 = self._get_filtered_chunk(ich)
filtered_chunk[:, pos : pos + end0 - start0] = filtered_chunk0[:, start0:end0]
pos += end0 - start0
def get_traces(self, start_frame=None, end_frame=None):
"""Grab filtered data iterating over chunks"""
if start_frame is None:
start_frame = 0
if end_frame is None:
end_frame = self.n_frames
if self._chunk_size is not None:
if end_frame - start_frame > self._chunk_size:
ich1 = int(start_frame / self._chunk_size)
ich2 = int((end_frame - 1) / self._chunk_size)
# write output to temp memmap for parallelization
with tempfile.NamedTemporaryFile(prefix=self._tmp_folder) as fp:
# create temp file
filtered_chunk = np.memmap(
fp,
dtype=self._dtype,
shape=(self.n_channels, int(end_frame - start_frame)),
mode="w+",
)
pos_list = []
start_list = []
end_list = []
pos = 0
for ich in range(ich1, ich2 + 1):
if ich == ich1:
start0 = start_frame - ich * self._chunk_size
else:
start0 = 0
if ich == ich2:
end0 = end_frame - ich * self._chunk_size
else:
end0 = self._chunk_size
pos_list.append(pos)
start_list.append(start0)
end_list.append(end0)
pos += end0 - start0
Parallel(n_jobs=self.n_jobs)(delayed(self._iterate_chunk)(filtered_chunk, pos, end0, start0, ich)
for pos, start0, end0, ich in zip(
tqdm(pos_list, disable=not(self.use_tqdm)), start_list, end_list, range(ich1, ich2 + 1)
)
)
if self.flat:
return filtered_chunk.astype(self._dtype).flatten()
else:
return filtered_chunk.astype(self._dtype)
filtered_chunk = self.filter_chunk(start_frame=0, end_frame=end_frame)
if self.flat:
return filtered_chunk.astype(self._dtype).flatten()
else:
return filtered_chunk.astype(self._dtype)
class SpectralGateNonStationary(SpectralGate):
def __init__(
self,
y,
sr,
chunk_size,
padding,
n_fft,
win_length,
hop_length,
time_constant_s,
freq_mask_smooth_hz,
time_mask_smooth_ms,
thresh_n_mult_nonstationary,
sigmoid_slope_nonstationary,
tmp_folder,
prop_decrease,
use_tqdm,
n_jobs
):
self._thresh_n_mult_nonstationary = thresh_n_mult_nonstationary
self._sigmoid_slope_nonstationary = sigmoid_slope_nonstationary
super().__init__(
y=y,
sr=sr,
chunk_size=chunk_size,
padding=padding,
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length,
time_constant_s=time_constant_s,
freq_mask_smooth_hz=freq_mask_smooth_hz,
time_mask_smooth_ms=time_mask_smooth_ms,
tmp_folder=tmp_folder,
prop_decrease=prop_decrease,
use_tqdm=use_tqdm,
n_jobs = n_jobs
)
def spectral_gating_nonstationary(self, chunk):
"""non-stationary version of spectral gating"""
denoised_channels = np.zeros(chunk.shape, chunk.dtype)
for ci, channel in enumerate(chunk):
sig_stft = librosa.stft(
(channel),
n_fft=self._n_fft,
hop_length=self._hop_length,
win_length=self._win_length,
)
# get abs of signal stft
abs_sig_stft = np.abs(sig_stft)
# get the smoothed mean of the signal
sig_stft_smooth = get_time_smoothed_representation(
abs_sig_stft,
self.sr,
self._hop_length,
time_constant_s=self._time_constant_s,
)
# get the number of X above the mean the signal is
sig_mult_above_thresh = (abs_sig_stft - sig_stft_smooth) / sig_stft_smooth
# mask based on sigmoid
sig_mask = sigmoid(
sig_mult_above_thresh, -self._thresh_n_mult_nonstationary, self._sigmoid_slope_nonstationary
)
if self.smooth_mask:
# convolve the mask with a smoothing filter
sig_mask = scipy.signal.fftconvolve(
sig_mask, self._smoothing_filter, mode="same"
)
sig_mask = sig_mask * self._prop_decrease + np.ones(np.shape(sig_mask)) * (1.0 - self._prop_decrease)
# multiply signal with mask
sig_stft_denoised = sig_stft * sig_mask
# invert/recover the signal
denoised_signal = librosa.istft(
sig_stft_denoised,
hop_length=self._hop_length,
win_length=self._win_length,
)
denoised_channels[ci, : len(denoised_signal)] = denoised_signal
return denoised_channels
def _do_filter(self, chunk):
"""Do the actual filtering"""
chunk_filtered = self.spectral_gating_nonstationary(chunk)
return chunk_filtered
class SpectralGateStationary(SpectralGate):
def __init__(
self,
y,
sr,
y_noise,
n_std_thresh_stationary,
chunk_size,
clip_noise_stationary,
padding,
n_fft,
win_length,
hop_length,
time_constant_s,
freq_mask_smooth_hz,
time_mask_smooth_ms,
tmp_folder,
prop_decrease,
use_tqdm,
n_jobs
):
super().__init__(
y=y,
sr=sr,
chunk_size=chunk_size,
padding=padding,
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length,
time_constant_s=time_constant_s,
freq_mask_smooth_hz=freq_mask_smooth_hz,
time_mask_smooth_ms=time_mask_smooth_ms,
tmp_folder=tmp_folder,
prop_decrease=prop_decrease,
use_tqdm=use_tqdm,
n_jobs = n_jobs
)
self.n_std_thresh_stationary = n_std_thresh_stationary
if y_noise is None:
self.y_noise = self.y
else:
y_noise = np.array(y_noise)
# reshape data to (#channels, #frames)
if len(y_noise.shape) == 1:
self.y_noise = np.expand_dims(y_noise, 0)
elif len(y.shape) > 2:
raise ValueError("Waveform must be in shape (# frames, # channels)")
else:
self.y_noise = y_noise
# collapse y_noise to one channel
self.y_noise = np.mean(self.y_noise, axis=0)
if clip_noise_stationary:
self.y_noise = self.y_noise[:chunk_size]
# calculate statistics over y_noise
abs_noise_stft = np.abs(librosa.stft(
(self.y_noise),
n_fft=self._n_fft,
hop_length=self._hop_length,
win_length=self._win_length,
))
noise_stft_db = _amp_to_db(abs_noise_stft)
self.mean_freq_noise = np.mean(noise_stft_db, axis=1)
self.std_freq_noise = np.std(noise_stft_db, axis=1)
self.noise_thresh = self.mean_freq_noise + self.std_freq_noise * self.n_std_thresh_stationary
def spectral_gating_stationary(self, chunk):
"""non-stationary version of spectral gating"""
denoised_channels = np.zeros(chunk.shape, chunk.dtype)
for ci, channel in enumerate(chunk):
sig_stft = librosa.stft(
(channel),
n_fft=self._n_fft,
hop_length=self._hop_length,
win_length=self._win_length,
)
# spectrogram of signal in dB
sig_stft_db = _amp_to_db(np.abs(sig_stft))
# calculate the threshold for each frequency/time bin
db_thresh = np.repeat(
np.reshape(self.noise_thresh, [1, len(self.mean_freq_noise)]),
np.shape(sig_stft_db)[1],
axis=0,
).T
# mask if the signal is above the threshold
sig_mask = sig_stft_db > db_thresh
sig_mask = sig_mask * self._prop_decrease + np.ones(np.shape(sig_mask)) * (1.0 - self._prop_decrease)
if self.smooth_mask:
# convolve the mask with a smoothing filter
sig_mask = scipy.signal.fftconvolve(
sig_mask, self._smoothing_filter, mode="same"
)
# multiply signal with mask
sig_stft_denoised = sig_stft * sig_mask
# invert/recover the signal
denoised_signal = librosa.istft(
sig_stft_denoised,
hop_length=self._hop_length,
win_length=self._win_length,
)
denoised_channels[ci, : len(denoised_signal)] = denoised_signal
return denoised_channels
def _do_filter(self, chunk):
"""Do the actual filtering"""
chunk_filtered = self.spectral_gating_stationary(chunk)
return chunk_filtered
def reduce_noise(
y,
sr,
stationary=False,
y_noise = None,
prop_decrease = 1.0,
time_constant_s=2.0,
freq_mask_smooth_hz=500,
time_mask_smooth_ms=50,
thresh_n_mult_nonstationary=2,
sigmoid_slope_nonstationary=10,
n_std_thresh_stationary = 1.5,
tmp_folder=None,
chunk_size=600000,
padding=30000,
n_fft=1024,
win_length=None,
hop_length=None,
clip_noise_stationary = True,
use_tqdm=False,
n_jobs = 1
):
"""
Reduce noise via spectral gating.
Parameters
----------
y : np.ndarray [shape=(# frames,) or (# channels, # frames)], real-valued
input signal
sr : int
sample rate of input signal / noise signal
y_noise : np.ndarray [shape=(# frames,) or (# channels, # frames)], real-valued
noise signal to compute statistics over (only for non-stationary noise reduction).
stationary : bool, optional
Whether to perform stationary, or non-stationary noise reduction, by default False
prop_decrease : float, optional
The proportion to reduce the noise by (1.0 = 100%), by default 1.0
time_constant_s : float, optional
The time constant, in seconds, to compute the noise floor in the non-stationary
algorithm, by default 2.0
freq_mask_smooth_hz : int, optional
The frequency range to smooth the mask over in Hz, by default 500
time_mask_smooth_ms : int, optional
The time range to smooth the mask over in milliseconds, by default 50
thresh_n_mult_nonstationary : int, optional
Only used in nonstationary noise reduction., by default 1
sigmoid_slope_nonstationary : int, optional
Only used in nonstationary noise reduction., by default 10
n_std_thresh_stationary : int, optional
Number of standard deviations above mean to place the threshold between
signal and noise., by default 1.5
tmp_folder : [type], optional
Temp folder to write waveform to during parallel processing. Defaults to
default temp folder for python., by default None
chunk_size : int, optional
Size of signal chunks to reduce noise over. Larger sizes
will take more space in memory, smaller sizes can take longer to compute.
, by default 60000
padding : int, optional
How much to pad each chunk of signal by. Larger pads are
needed for larger time constants., by default 30000
n_fft : int, optional
length of the windowed signal after padding with zeros.
The number of rows in the STFT matrix ``D`` is ``(1 + n_fft/2)``.
The default value, ``n_fft=2048`` samples, corresponds to a physical
duration of 93 milliseconds at a sample rate of 22050 Hz, i.e. the
default sample rate in librosa. This value is well adapted for music
signals. However, in speech processing, the recommended value is 512,
corresponding to 23 milliseconds at a sample rate of 22050 Hz.
In any case, we recommend setting ``n_fft`` to a power of two for
optimizing the speed of the fast Fourier transform (FFT) algorithm., by default 1024
win_length : [type], optional
Each frame of audio is windowed by ``window`` of length ``win_length``
and then padded with zeros to match ``n_fft``.
Smaller values improve the temporal resolution of the STFT (i.e. the
ability to discriminate impulses that are closely spaced in time)
at the expense of frequency resolution (i.e. the ability to discriminate
pure tones that are closely spaced in frequency). This effect is known
as the time-frequency localization trade-off and needs to be adjusted
according to the properties of the input signal ``y``.
If unspecified, defaults to ``win_length = n_fft``., by default None
hop_length : [type], optional
number of audio samples between adjacent STFT columns.
Smaller values increase the number of columns in ``D`` without
affecting the frequency resolution of the STFT.
If unspecified, defaults to ``win_length // 4`` (see below)., by default None
use_tqdm : bool, optional
Whether to show tqdm progress bar, by default False
n_jobs : int, optional
Number of parallel jobs to run. Set at -1 to use all CPU cores, by default 1
"""
if stationary:
sg = SpectralGateStationary(
y=y,
sr=sr,
y_noise = y_noise,
prop_decrease =prop_decrease,
n_std_thresh_stationary = n_std_thresh_stationary,
chunk_size=chunk_size,
clip_noise_stationary = clip_noise_stationary,
padding=padding,
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length,
time_constant_s=time_constant_s,
freq_mask_smooth_hz=freq_mask_smooth_hz,
time_mask_smooth_ms=time_mask_smooth_ms,
tmp_folder=tmp_folder,
use_tqdm=use_tqdm,
n_jobs = n_jobs
)
else:
sg = SpectralGateNonStationary(
y=y,
sr=sr,
chunk_size=chunk_size,
padding=padding,
prop_decrease =prop_decrease,
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length,
time_constant_s=time_constant_s,
freq_mask_smooth_hz=freq_mask_smooth_hz,
time_mask_smooth_ms=time_mask_smooth_ms,
thresh_n_mult_nonstationary=thresh_n_mult_nonstationary,
sigmoid_slope_nonstationary=sigmoid_slope_nonstationary,
tmp_folder=tmp_folder,
use_tqdm=use_tqdm,
n_jobs = n_jobs
)
return sg.get_traces()
def _amp_to_db(x):
return librosa.core.amplitude_to_db(x, ref=1.0, amin=1e-20, top_db=80.0)
def _db_to_amp(x,):
return librosa.core.db_to_amplitude(x, ref=1.0)