def bandpass_filter(s,
sample_rate,
low_freq,
high_freq,
filter_order=5,
rescale=False):
"""
Bandpass filter a signal s.
s: the signal
sample_rate: the sample rate in Hz of the signal
low_freq: the lower cutoff frequency
upper_freq: the upper cutoff frequency
filter_order: the order of the filter...
Returns the bandpass filtered signal s.
"""
#create a butterworth filter
nyq = sample_rate / 2.0
f = np.array([low_freq, high_freq]) / nyq
b, a = filter_design.butter(filter_order, f, btype='bandpass')
#filter the signal
filtered_s = filtfilt(b, a, s)
if rescale:
#rescale filtered signal
filtered_s /= filtered_s.max()
filtered_s *= s.max()
return filtered_s
def lowpass(data, in_t, cutoff, order=4):
"""
data: vector of data
in_t: sample times
cutoff: cutoff period in the same units as in_t
returns vector same as data, but with high frequencies removed
"""
# Step 1: Determine dt from data (complicated due to TRIM time series precision legacy)
mean_dt = mean(diff(in_t))
# print 'Data Time Step: %8.4f'%mean_dt
# test size of data for vectors only
Wn = mean_dt / cutoff # 90 = half # of minutes in 3 hours.
# print 'Using %1ith Order Butterworth Filter, with %g time-unit cutoff (Wn=%12.6f).'%(order,2*cutoff,Wn)
B, A = butter(order, Wn)
data_filtered = filtfilt(B, A, data)
return data_filtered
def lowpass(data,
in_t=None,
cutoff=None,
order=4,
dt=None,
axis=-1,
causal=False):
"""
data: vector of data
in_t: sample times
cutoff: cutoff period in the same units as in_t
returns vector same as data, but with high frequencies removed
"""
# Step 1: Determine dt from data or from user if specified
if dt is None:
dt = np.median(np.diff(in_t))
dt = float(dt) # make sure it's not an int
cutoff = float(cutoff)
Wn = dt / cutoff
B, A = butter(order, Wn)
if not causal:
# scipy filtfilt triggers some warning message about tuple
# indices.
with warnings.catch_warnings():
warnings.simplefilter("ignore")
data_filtered = filtfilt(B, A, data, axis=axis)
else:
data_filtered = lfilter(B, A, data, axis=axis)
return data_filtered
def lowpass(data, in_t, cutoff, order=4):
"""
data: vector of data
in_t: sample times
cutoff: cutoff period in the same units as in_t
returns vector same as data, but with high frequencies removed
"""
# Step 1: Determine dt from data (complicated due to TRIM time series precision legacy)
mean_dt = mean(diff(in_t))
# print 'Data Time Step: %8.4f'%mean_dt
# test size of data for vectors only
Wn = mean_dt / cutoff # 90 = half # of minutes in 3 hours.
# print 'Using %1ith Order Butterworth Filter, with %g time-unit cutoff (Wn=%12.6f).'%(order,2*cutoff,Wn)
B, A = butter(order, Wn)
data_filtered = filtfilt(B, A, data)
return data_filtered
def bandpass_filter(s, sample_rate, low_freq, high_freq, filter_order=5, rescale=False):
"""
Bandpass filter a signal s.
s: the signal
sample_rate: the sample rate in Hz of the signal
low_freq: the lower cutoff frequency
upper_freq: the upper cutoff frequency
filter_order: the order of the filter...
Returns the bandpass filtered signal s.
"""
#create a butterworth filter
nyq = sample_rate / 2.0
f = np.array([low_freq, high_freq]) / nyq
b,a = filter_design.butter(filter_order, f, btype='bandpass')
#filter the signal
filtered_s = filtfilt(b, a, s)
if rescale:
#rescale filtered signal
filtered_s /= filtered_s.max()
filtered_s *= s.max()
return filtered_s
def highpass_filter(s,
sample_rate,
cutoff_freq,
filter_order=5,
rescale=False):
"""
Highpass filter a signal s, with sample rate sample_rate.
s: the signal
sample_rate: the sample rate in Hz of the signal
cutoff_freq: the cutoff frequency of the filter
filter_order: the order of the filter...
Returns the low-pass filtered signal s.
"""
#create a butterworth filter
nyq = sample_rate / 2.0
b, a = filter_design.butter(filter_order, cutoff_freq / nyq, btype='high')
#filter the signal
filtered_s = filtfilt(b, a, s)
if rescale:
#rescale filtered signal
filtered_s /= filtered_s.max()
filtered_s *= s.max()
return filtered_s
def lpFiltSpec(fp, fs, nyqf):
nyqf = float(nyqf)
ord, wn = filter_design.buttord(fp / nyqf, fs / nyqf, 1, 12, analog=0)
b, a = butter(ord, wn, btype="lowpass", analog=0, output='ba')
w, h = freqz(b, a)
fig = _plt.figure()
_plt.plot(w * (nyqf / _N.pi), _N.abs(h))
def highpass_filter(s, sample_rate, cutoff_freq, filter_order=5, rescale=False):
"""
Highpass filter a signal s, with sample rate sample_rate.
s: the signal
sample_rate: the sample rate in Hz of the signal
cutoff_freq: the cutoff frequency of the filter
filter_order: the order of the filter...
Returns the low-pass filtered signal s.
"""
#create a butterworth filter
nyq = sample_rate / 2.0
b,a = filter_design.butter(filter_order, cutoff_freq / nyq, btype='high')
#filter the signal
filtered_s = filtfilt(b, a, s)
if rescale:
#rescale filtered signal
filtered_s /= filtered_s.max()
filtered_s *= s.max()
return filtered_s
def butter_bandpass_filter_two(data, lowcut, highcut, fs, order=5):
nyq = 0.5 * fs
low = lowcut / nyq
high = highcut / nyq
b, a = butter(order, [low, high], btype='band')
y = lfilter(b, a, data)
return y
def lowpass(data,
in_t=None,
cutoff=None,
order=4,
dt=None,
axis=-1,
causal=False):
"""
data: vector of data
in_t: sample times
cutoff: cutoff period in the same units as in_t
returns vector same as data, but with high frequencies removed
"""
# Step 1: Determine dt from data
dt = dt or np.median(np.diff(in_t))
dt = float(dt)
cutoff = float(cutoff)
Wn = dt / cutoff
B, A = butter(order, Wn)
if not causal:
data_filtered = filtfilt(B, A, data, axis=axis)
else:
data_filtered = lfilter(B, A, data, axis=axis)
return data_filtered
def envelope(signal, fs, env_cutoff=16., env_order=4., full_rect=False):
'''Extracts the amplitude envelope from a signal using rectification and low-pass filtering
Parameters
----------
signal : array
The signal to extract the amplitude envelope from
fs : scalar
The sampling frequency
env_cutoff : scalar
The envelope cutoff frequency [default=16 Hz]
env_order : scalar
The order of the envelope filter [must be even number; default=6]
full_rect : bool
True for full-wave rectification [default=half-wave rectification]
Returns
-------
env : array
The amplitude envelope of the input signal
'''
signal = signal - np.mean(signal) # Remove DC component
if full_rect:
rect = np.absolute(signal)
else:
rect = np.maximum(signal,0)
env_b,env_a = filters.butter(env_order/2.,np.float32(env_cutoff)/(np.float32(fs)/2.))
env = filtfilt(env_b,env_a,rect)
return env
def bpFilt(fpL, fpH, fsL, fsH, nyqf, y):
nyqf = float(nyqf)
ord, wn = filter_design.buttord((fpL / nyqf, fpH / nyqf),
(fsL / nyqf, fsH / nyqf),
1,
12,
analog=0)
b, a = butter(ord, wn, btype="bandpass", analog=0, output='ba')
fy = filtfilt(b, a, y)
return fy
def band_pass_filter(self, wavFile):
rate, sound_samples = read(wavFile) # f , fs
sound_samples = np.float64(sound_samples / 32768.0)
n = 7
beginFreq = 700.0 / (rate / 2)
endFreq = 12000.0 / (rate / 2)
[b, a] = butter(n, [beginFreq, endFreq], 'bandpass')
filtered = lfilter(b, a, sound_samples)
filtered = np.int16(filtered * 32768 * 10)
write(self.__filteredFile, rate, filtered)
return self.__filteredFile
def bpFiltSpec(fpL, fpH, fsL, fsH, nyqf):
nyqf = float(nyqf)
ord, wn = filter_design.buttord((fpL / nyqf, fpH / nyqf),
(fsL / nyqf, fsH / nyqf),
1,
12,
analog=0)
b, a = butter(ord, wn, btype="bandpass", analog=0, output='ba')
w, h = freqz(b, a)
fig = _plt.figure()
_plt.plot(w * (nyqf / _N.pi), _N.abs(h))
def lowpass(signal, order, cutoff, sr):
b, a = butter(order, cutoff / (sr / 2))
filteredSignal = nans(shape(signal))
if len(signal) <= 3 * max(len(a), len(b)):
print "signal too short to be lowpassed: " + str(len(signal))
return nans(shape(signal))
for i in range(shape(signal)[1]):
filteredSignal[:, i] = filtfilt(b, a, signal[:, i])
return filteredSignal
def low_filter(self, wavFile):
rate, sound_samples = read(wavFile)
sound_samples = np.float64(sound_samples / 32768.0)
pass_freq = 0.2
stop_freq = 0.3
pass_gain = 0.5 # permissible loss (ripple) in passband (dB)
stop_gain = 10.0 # attenuation required in stopband (dB)
ord, wn = buttord(pass_freq, stop_freq, pass_gain, stop_gain)
b, a = butter(ord, wn, btype='low')
filtered = lfilter(b, a, sound_samples)
filtered = np.int16(filtered * 32768 * 10)
write(self.__filteredFile, rate, filtered)
return self.__filteredFile
def butter_bandpass_filter(data, lowcut, highcut, fs, order=5):
"""
Runs a Butterworth bandpass filter on sound data
:param data: sound data to filter
:param lowcut: the lowest frequency to accept
:param highcut: the highest frequency to accept
:param fs: sampling frequency
:param order: the order of the filter
"""
nyq = 0.5 * fs
low = lowcut / nyq
high = highcut / nyq
b, a = butter(order, [low, high], btype='band')
y = lfilter(b, a, data)
return y
def butter_bandpass_filter(data, lowcut, highcut, fs, order=5):
"""
Runs a Butterworth bandpass filter on sound data
:param data: sound data to filter
:param lowcut: the lowest frequency to accept
:param highcut: the highest frequency to accept
:param fs: sampling frequency
:param order: the order of the filter
"""
nyq = 0.5 * fs
low = lowcut / nyq
high = highcut / nyq
b, a = butter(order, [low, high], btype='band')
y = lfilter(b, a, data)
return y
def butter_pass(cutoff, fs, order=5, btype="low"):
nyq = 0.5 * fs
normal_cutoff = cutoff / nyq
b, a = butter(order, normal_cutoff, btype=btype, analog=False)
return b, a
def butter_bp(lo=0, hi=0, Fs=2.0, ord=3):
# note: "lo" corresponds to highpass cutoff
# "hi" corresponds to lowpass cutoff
freqs, btype = _bandpass_params(lo, hi)
return fdesign.butter(ord, 2 * freqs / Fs, btype=btype)
def butter_bandpass(lowcut, highcut, fs, order=9):
nyq = 0.5 * fs
low = lowcut / nyq
high = highcut / nyq
b, a = butter(order, [low, high], btype='band')
return b, a
format(data_wav.shape[0] / fs_wav)) duration = (data_wav.shape[0] / fs_wav) print(duration, " Duration") count_minutes = int(duration/60) print(count_minutes, " minutes") bpm=[] for j in range(1,count_minutes+1): print(j,"th iteration") data_wav_norm = copy.deepcopy(data_wav) data_wav_norm = data_wav_norm[(j-1)*60 * fs_wav: (j)*60 * fs_wav] time_wav = np.arange(0, len(data_wav_norm)) / fs_wav # Applying Low-pass Butter filter b, a = butter(3, 0.9/(fs_wav*0.5), btype = 'low', analog=False) filtered = lfilter(b, a, data_wav_norm) data_wav_lp = filtered print(len(data_wav_lp), "Length") data_wav_lp = data_wav_lp/ (2**15) #Applying hilbert Transformation to get envelope of the signal analytic_signal = hilbert(data_wav_lp) amplitude_envelope = np.abs(analytic_signal) instantaneous_phase = np.unwrap(np.angle(analytic_signal)) #getting Properties of the envelope peaks, prop = find_peaks(amplitude_envelope,height=[0,0.2], distance=5000) peak_graph = prop["peak_heights"]
def f0(sig, fs, noisegate=15):
'''Estimates the fundamental frequency of a signal
Returns an array of estimated instantaneous fundamental frequency
(F0) values based on zeros crossings in the input signal.
Parameters
----------
sig : array
The input signal.
fs : array
The sampling frequency.
Returns
-------
y : array
The pitch track.
'''
# Low-pass at 270 Hz (above most F0's)
b, a = filters.butter(4, 270. / (fs / 2.))
fsig = filtfilt(b, a, sig)
b, a = filters.butter(4, 60. / (fs / 2.), btype='high')
fsig = filtfilt(b, a, fsig)
# Get zero crossings
zc = np.array(
np.where(np.sign(fsig[1:]) != np.sign(fsig[:-1]))).transpose()[:, 0]
# Compute period at each crossing
# (actually half-period, since a full period is 2 zero crossings)
zc2 = np.concatenate((zc, np.zeros((1))))
zc3 = np.concatenate((np.zeros((1)), zc))
p = zc2 - zc3
p[-1] = p[-2]
# Create new array of periods @ fs
pfs = np.zeros(fsig.shape)
# Resample period data (aliased)
for k in range(zc.shape[0] - 1):
pfs[zc[k]:zc[k + 1]] = p[k]
pfs[0:zc[0]] = p[0]
pfs[zc[-1]:] = p[-2]
# Convert instantaneous period data to instantaneous frequency data
ps = ((pfs / (fs / 1000.)) / 1000.)
# The 2 here corrects for the half-period issue above
f = (1. / ps) / 2.
# Smooth the F0 track
b, a = filters.butter(1, 16. / (fs / 2.))
f = filtfilt(b, a, f)
# Voicing
env = envelope(fsig, fs)
env = env / np.max(np.abs(env))
# Noise gate
env[20 * np.log10(env) < -noisegate] = 0
env[env > 0] = 1
# env2 = env
# vonsets = array(where(sign(env[1:]) < sign(env[:-1]))).transpose()[:,0]
# print vonsets
# print vonsets + vdelay - 1
# skip = True
# for k in range(vonsets.shape[0]):
# if (skip):
# env2[vonsets[k]:vonsets[k] + vdelay - 1] = 0
# skip = False
# else:
# skip = True
# Smooth the transitions
b, a = filters.butter(1, 16. / (fs / 2.))
envf = filtfilt(b, a, env)
return f * np.maximum(envf, 0)
if((n-k)>= len(int_resp)):
temp += 0
else:
temp += data_buffer[k%64]*int_resp[n-k]
data_dump[n%64] = temp;
f = open('sample_vishwa.dat', 'r')
data = []
for i in f.readlines():
data.append(int(i))
data_test = zeros(len(data))
data_test[0] = 1
order, Wn = fd.buttord(0.2, 0.3, 1, 10)
print order, Wn
b, a = fd.butter(order, Wn)
response = sg.lfilter(b, a, data_test)[:256]
int_resp = []
resp_file = open('impulse_response_integer.dat', 'r')
for i in resp_file.readlines():
int_resp.append(int(i))
#op1 = conv(data[:1024], int_resp)
f.close()
op2 = range(1024)
conv(data)
op1 = convolve(int_resp, data)
clf()
plot(op1, 'r')
plot(op2)
show()
def hpFilt(fp, fs, nyqf, y, disp=False):
nyqf = float(nyqf)
ord, wn = filter_design.buttord(fp / nyqf, fs / nyqf, 1, 12, analog=0)
b, a = butter(ord, wn, btype="highpass", analog=0, output='ba')
fy = filtfilt(b, a, y)
return fy
def butterworth(ts, cutoff_period=None, cutoff_frequency=None, order=4):
""" low-pass butterworth-squared filter on a regular time series.
Parameters
-----------
ts : :class:`DataFrame <pandas:pandas.DataFrame>`
Must be one or two dimensional, and regular.
order: int ,optional
The default is 4.
cutoff_frequency: float,optional
Cutoff frequency expressed as a ratio with Nyquist frequency,
should within the range (0,1). For a discretely sampled system,
the Nyquist frequency is the fastest frequency that can be resolved by that
sampling, which is half the sampling frequency. For example, if the sampling frequency
is 1 sample/1 hour, the Nyquist frequency is 1 sample/2 hours. If we want a
36 hour cutoff period, the frequency is 1/36 or 0.0278 cycles per hour.
Hence the cutoff frequency argument used here would be
0.0278/0.5 = 0.056.
cutoff_period : string or :ref:`time_interval<time_intervals>`
Period corresponding to cutoff frequency. If input as a string, it must
be convertible to a regular interval using the same rules as a pandas frequency..
cutoff_frequency and cutoff_period can't be specified at the same time.
Returns
-------
result :
A new regular time series with the same interval as ts.
Raise
--------
ValueError
If input order is not even, or input timeseries is not regular,
or neither cutoff_period and cutoff_frequency is given while input
time series interval is not 15min or 1 hour, or cutoff_period and cutoff_frequency
are given at the same time.
"""
if (order % 2):
raise ValueError("only even order is accepted")
#if not ts.is_regular():
# raise ValueError("Only regular time series can be filtered.")
freq = ts.index.freq
# if (not (interval in _butterworth_interval)) and (cutoff_period is None) and (cutoff_frequency is None):
# raise ValueError("time interval is not supported by butterworth if no cuttoff period/frequency given.")
if (not (cutoff_frequency is None)) and (not (cutoff_period is None)):
raise ValueError(
"cutoff_frequency and cutoff_period can't be specified simultaneously"
)
if (cutoff_frequency is None) and (cutoff_period is None):
raise ValueError(
"Either cutoff_frequency or cutoff_period must be given")
cf = cutoff_frequency
if (cf is None):
if (not (cutoff_period is None)):
cutoff_period = pd.tseries.frequencies.to_offset(cutoff_period)
cf = 2. * freq / cutoff_period
else:
cf = butterworth_cutoff_frequencies[interval]
## get butter filter coefficients.
[b, a] = butter(order / 2, cf)
d2 = filtfilt(b, a, ts.values, axis=0, padlen=90)
out = ts.copy(deep=True)
out[:] = d2
# prop={}
# for key,val in ts.props.items():
# prop[key]=val
# prop[TIMESTAMP]=INST
# prop[AGGREGATION]=INDIVIDUAL
# time_interval
return out
def butterworth(ts, order=4, cutoff_period=None, cutoff_frequency=None):
""" low-pass butterworth-squared filter on a regular time series.
Parameters
-----------
ts : :class:`~vtools.data.timeseries.TimeSeries`
Must has data of one dimension, and regular.
order: int ,optional
The default is 4.
cutoff_frequency: float,optional
Cutoff frequency expressed as a ratio of a Nyquist frequency,
should within the range (0,1). For example, if the sampling frequency
is 1 hour, the Nyquist frequency is 1 sample/2 hours. If we want a
36 hour cutoff period, the frequency is 1/36 or 0.0278 cycles per hour.
Hence the cutoff frequency argument used here would be
0.0278/0.5 = 0.056.
cutoff_period : string or :ref:`time_interval<time_intervals>`
Period of cutting off frequency. If input as a string, it must
be convertible to :ref:`Time interval<time_intervals>`.
cutoff_frequency and cutoff_period can't be specified at the same time.
Returns
-------
result : :class:`~vtools.data.timeseries.TimeSeries`
A new regular time series with the same interval of ts.
Raise
--------
ValueError
If input order is not even, or input timeseries is not regular,
or neither cutoff_period and cutoff_frequency is given while input
time series interval is not 15min or 1 hour, or cutoff_period and cutoff_frequency
are given at the same time.
"""
if (order % 2):
raise ValueError("only even order is accepted")
if not ts.is_regular():
raise ValueError("Only regular time series can be filtered.")
interval = ts.interval
if (not (interval in _butterworth_interval)) and (
cutoff_period is None) and (cutoff_frequency is None):
raise ValueError(
"time interval is not supported by butterworth if no cuttoff period/frequency given."
)
if (not (cutoff_frequency is None)) and (not (cutoff_period is None)):
raise ValueError(
"cutoff_frequency and cutoff_period can't be specified simultaneously"
)
cf = cutoff_frequency
if (cf is None):
if (not (cutoff_period is None)):
## convert it to ticks
if not (is_interval(cutoff_period)):
cutoff_period = parse_interval(cutoff_period)
cutoff_frequency_in_ticks = 1.0 / float(ticks(cutoff_period))
nyquist_frequency = 0.5 / float(ticks(interval))
cf = cutoff_frequency_in_ticks / nyquist_frequency
else:
cf = butterworth_cutoff_frequencies[interval]
## get butter filter coefficients.
[b, a] = butter(order / 2, cf)
# d1=lfilter(b, a, ts.data,0)
# d1=d1[len(d1)::-1]
# d2=lfilter(b,a,d1,0)
# d2=d2[len(d2)::-1]
d2 = filtfilt(b, a, ts.data, axis=0)
prop = {}
for key, val in ts.props.items():
prop[key] = val
prop[TIMESTAMP] = INST
prop[AGGREGATION] = INDIVIDUAL
time_interval
return rts(d2, ts.start, ts.interval, prop)
def vocoder_overlap(signal, fs, channel_n, channel_width, flo, fhi):
'''Prototype vocoder where channel width is independent of channel spacing
That is, channels are not necessarily contiguous
The channel_width parameter is in octaves, and channel cfs are
logarithmically spaced from flo to fhi
E.g., the following set of parameters would yield contiguous bands from 88 to 11314 Hz:
psylab.signal.vocoder_overlap(sig,fs,6,1,125,8000)
vocoder_overlap(signal, fs, channel_n, channel_width, flo, fhi)
'''
signal = signal - np.mean(signal)
cfs = np.float32(np.round(np.linspace(flo, fhi, channel_n)))
cfs = logspace(flo, fhi, channel_n)
cw = np.float32(channel_width / 2.)
nyq = np.float32(fs / 2.)
envfilter = 400.
print("Channel width: {:} Oct".format(channel_width))
noisecarrier = np.random.randn(len(signal))
noisecarrier = noisecarrier / max(np.abs(noisecarrier))
summed_carriers = np.zeros(len(signal))
for cf in cfs:
lo = np.round(cf * (2.**-cw))
hi = np.round(cf * (2.**cw))
print(" lo {:}; cf {:}; hi {:}".format(lo, cf, hi))
[b_band_hp, a_band_hp] = filters.butter(3, (lo / nyq), btype='high')
[b_band_lp, a_band_lp] = filters.butter(3, (hi / nyq))
[b_wind_hp, a_wind_hp] = filters.butter(1, (cf / nyq), btype='high')
[b_wind_lp, a_wind_lp] = filters.butter(1, (cf / nyq))
[b_env, a_env] = filters.butter(2,
min((.5 * (hi - lo)), envfilter) / nyq)
# Filter signal into sub-band
Sig_band = lfilter(b_band_hp, a_band_hp, signal)
Sig_band = lfilter(b_band_lp, a_band_lp, Sig_band)
# Apply window to shape band
Sig_band = lfilter(b_wind_hp, a_wind_hp, Sig_band)
Sig_band = lfilter(b_wind_lp, a_wind_lp, Sig_band)
# Extract envelope
rms_Sig_band = np.sqrt(np.mean(Sig_band**2))
Sig_env_band = lfilter(b_env, a_env, np.maximum(Sig_band, 0))
# Prefilter, modulate carrier
Mod_carrier = filtfilt(b_band_hp, a_band_hp, noisecarrier)
Mod_carrier = filtfilt(b_band_lp, a_band_lp,
Mod_carrier) * Sig_env_band
# Post filter
Mod_carrier_filt = lfilter(b_band_hp, a_band_hp, Mod_carrier)
Mod_carrier_filt = lfilter(b_band_lp, a_band_lp, Mod_carrier_filt)
summed_carriers += Mod_carrier / np.sqrt(np.mean(Mod_carrier**
2)) * rms_Sig_band
return summed_carriers * (np.sqrt(np.mean(signal**2)) /
np.sqrt(np.mean(summed_carriers**2)))
def vocoder(signal, fs, channels, inlo, inhi, **kwargs):
'''Implements an envelope vocoder
Vocodes the input signal using specified parameters.
You can specify a different frequency range for both input
(analysis) and output, for a given number of channels (the output
frequency range values default to input values). Both input and output
channels are distributed contiguously on a log scale across the
respective frequency ranges. Carriers can be pure tones (at the
arithmetic center of each band) or noise bands (which are as wide as
the output channel). The envelope cutoff frequency defaults to 400 Hz,
but is never more than half the output-channel bandwidth.
Parameters
----------
signal : array
The input signal
fs : scalar
The sampling frequency
channels : scalar
The number of vocoder channels
inlo : scalar
Low-side (start) frequency of the analysis channels
inhi : scalar
High-side (end) frequency of the analysis channels
Kwargs
------
outlo : scalar
Low-side (start) frequency of the output channels [ default = inlo ]
outhi : scalar
High-side (end) frequency of the output channels [ default = inhi ]
compression_ratio : scalar
Compression ration: input / output [ default = 1 ]
gate : scalar
A gate to apply to each envelope. Values below gate are set to 0 [ default = 0 ]
envfilter : scalar
Low-pass cutoff frequency of the envelope extraction filter [ default = 400 ]
noise : bool
False for sinusoidal carriers [ default ]
True for noise band carriers
sumchannels : bool
False to return a 2-d array in which each output channel is a column
True to return a 1-d array containing the summed output channels. The rms
will be equated to the rms of the input [ default ]
order : int
The filter order to use [ default = 3 ]
Returns
-------
y : array
The vocoded signal
Notes
-----
Depends on tone.py
'''
outlo = kwargs.get('outlo', inlo)
outhi = kwargs.get('outhi', inhi)
envfilter = kwargs.get('envfilter', 400)
noise = kwargs.get('noise', False)
sumchannels = kwargs.get('sumchannels', True)
ord = kwargs.get('order', 3)
compression_ratio = kwargs.get('compression_ratio', 1)
gate = kwargs.get('gate', None)
if noise:
noisecarrier = np.random.randn(len(signal))
noisecarrier = noisecarrier / max(np.abs(noisecarrier))
signal = signal - np.mean(signal)
nyq = np.float32(fs / 2.)
ininterval = np.log10(
np.float32(inhi) / np.float32(inlo)) / np.float32(channels)
outinterval = np.log10(
np.float32(outhi) / np.float32(outlo)) / np.float32(channels)
if sumchannels:
carriers = np.zeros(len(signal))
else:
carriers = np.zeros((len(signal), channels))
ret = np.zeros((len(signal), channels))
for i in range(channels):
# Estimate filters
finhi = np.float32(inlo) * 10.**(ininterval * (i + 1))
finlo = np.float32(inlo) * 10.**(ininterval * i)
fouthi = np.float32(outlo) * 10.**(outinterval * (i + 1))
foutlo = np.float32(outlo) * 10.**(outinterval * i)
fcarrier = .5 * (fouthi + foutlo)
[b_sub_hp, a_sub_hp] = filters.butter(ord, (finlo / nyq), btype='high')
[b_sub_lp, a_sub_lp] = filters.butter(ord, (finhi / nyq))
[b_env, a_env
] = filters.butter(2,
min((.5 * (fouthi - foutlo)), envfilter) / nyq)
[b_out_hp, a_out_hp] = filters.butter(ord, (foutlo / nyq),
btype='high')
[b_out_lp, a_out_lp] = filters.butter(ord, (fouthi / nyq))
## Filter input
Sig_sub = lfilter(b_sub_hp, a_sub_hp, signal)
Sig_sub = lfilter(b_sub_lp, a_sub_lp, Sig_sub)
# ret[:,i] = Sig_sub.copy()
rms_Sig_sub = np.sqrt(np.mean(Sig_sub**2))
Sig_env_sub = lfilter(b_env, a_env, np.maximum(Sig_sub, 0))
peak = Sig_env_sub.max()
Sig_env_sub /= compression_ratio
Sig_env_sub += peak - Sig_env_sub.max()
if gate is not None:
db = 20 * np.log10(Sig_env_sub / peak)
Sig_env_sub[db < gate] = 0
if noise:
Mod_carrier = filtfilt(b_out_hp, a_out_hp, noisecarrier)
Mod_carrier = filtfilt(b_out_lp, a_out_lp,
Mod_carrier) * Sig_env_sub
else:
Mod_carrier = tone(np.ones(len(signal)) * fcarrier, fs,
1) * Sig_env_sub
## Filter output
Mod_carrier_filt = lfilter(b_out_hp, a_out_hp, Mod_carrier)
Mod_carrier_filt = lfilter(b_out_lp, a_out_lp, Mod_carrier_filt)
if sumchannels:
carriers += Mod_carrier_filt / np.sqrt(np.mean(Mod_carrier_filt**
2)) * rms_Sig_sub
else:
carriers[:, i] = Mod_carrier_filt / np.sqrt(
np.mean(Mod_carrier_filt**2)) * rms_Sig_sub
# return ret
if sumchannels:
return carriers * (np.sqrt(np.mean(signal**2)) /
np.sqrt(np.mean(carriers**2)))
else:
return carriers
