def get_local_shimmer(sound,
min_time=0.,
max_time=0.,
pitch_floor=75.,
pitch_ceiling=600.,
period_floor=0.0001,
period_ceiling=0.02,
max_period_factor=1.3,
max_amplitude_factor=1.6):
"""
Function to calculate (local) shimmer from a periodic PointProcess.
:param (parselmouth.Sound) sound: sound waveform
:param (float) min_time: minimum time value considered for time range (t1, t2) (default: 0.)
:param (float) max_time: maximum time value considered for time range (t1, t2) (default: 0.)
NOTE: If max_time <= min_time, the entire time domain is considered
:param (float) pitch_floor: minimum pitch (default: 75.)
:param (float) pitch_ceiling: maximum pitch (default: 600.)
:param (float) period_floor: the shortest possible interval that will be used in the computation
of shimmer, in seconds (default: 0.0001)
:param (float) period_ceiling: the longest possible interval that will be used in the
computation of shimmer, in seconds (default: 0.02)
:param (float) max_period_factor: the largest possible difference between consecutive intervals
that will be used in the computation of shimmer (default: 1.3)
:param (float) max_amplitude_factor: maximum amplitude factor for shimmer (default: 1.6)
:return: value of (local) shimmer
"""
# Create a PointProcess object
point_process = call(sound, 'To PointProcess (periodic, cc)', pitch_floor,
pitch_ceiling)
local_shimmer = call([sound, point_process], 'Get shimmer (local)',
min_time, max_time, period_floor, period_ceiling,
max_period_factor, max_amplitude_factor)
return local_shimmer
def process(self):
"""cpp"""
try:
voice = self.args["voice"]
pitch_ceiling = self.args["Pitch Ceiling"]
pitch_floor = self.args["Pitch Floor"]
spectrum = voice.to_spectrum()
cepstrum = call(spectrum, "To PowerCepstrum")
# Call the provided pitch bounds functions
pitch_floor = self.args["Pitch Floor"]
pitch_ceiling = self.args["Pitch Ceiling"]
interpolation = self.args["interpolation"]
tilt_line_qeufrency_lower_bound = self.args[
"Tilt line qeufrency lower bound"]
tilt_line_qeufrency_upper_bound = self.args[
"Tilt line qeufrency upper bound"]
linetype = self.args["Line type"]
fitmethod = self.args["Fit method"]
cpp = call(
cepstrum,
"Get peak prominence",
pitch_floor,
pitch_ceiling,
interpolation,
tilt_line_qeufrency_lower_bound,
tilt_line_qeufrency_upper_bound,
linetype,
fitmethod,
)
return {"cpp": cpp}
except:
return {"cpp": "Measurement failed"}
def extract_syllable_intervals(file_name):
print("Extracting syllable intervals from '{}'...".format(file_name))
# Use Praat script to extract syllables
# For each file name, we first run the Praat script, passing the desired parameters
# This script was slightly adapted, as it used to take a directory as argument and loop
# over the audio files in that directory but now only takes a single file name
# and executes the algorithm for that file
# As described in the script file, these parameters are: 'Silence threshold (dB)',
# 'Minimum dip between peaks (dB)', 'Minimum pause duration', and the filename
objects = run_file('syllable_nuclei.praat', -25, 2, 0.3, file_name)
# The script selects two objects at the end, the Sound object and the TextGrid
# These two objects are returned in a list, and now we assign the second one to the variable 'textgrid'
textgrid = objects[1]
# Call the Praat command "Get number of points" to query the amount of points in the first tier
n = call(textgrid, "Get number of points", 1)
# Make a list that queries the time of the point in the TextGrid for all points 1 to n
# (through a Python 'list comprehension', in this case, but one could also repeatedly 'append')
syllable_nuclei = [
call(textgrid, "Get time of point", 1, i + 1) for i in range(n)
]
# Use NumPy to calculate intervals between the syllable nuclei
syllable_intervals = np.diff(syllable_nuclei)
return syllable_intervals
def to_lpc(self,
method:str,
prediction_order:int=16,
window_length:Real=0.025,
time_step:Real=0.005,
pre_emphasis_frequency:Real=50,
**kwargs:Any) -> pm.Data:
"""
Parameters
----------
method: str,
prediction_order: int 16,
window_length:real number, default 0.025,
time_step:real number, default 0.005,
pre_emphasis_frequency:real number, default 50,
kwargs: dict, optional,
"tolerance1", "tolerance2" for `method` "marple", both default 1.0e-6
"""
cmd = f"To LPC ({method})"
if method in ["autocorrelation", "covariance", "burg"]:
lpc = call(self, cmd, prediction_order, window_length, time_step, pre_emphasis_frequency)
elif method == "marple":
tolerance1 = kwargs.get("tolerance1", 1.0e-6)
tolerance2 = kwargs.get("tolerance2", 1.0e-6)
lpc = call(self, cmd, prediction_order, window_length, time_step, pre_emphasis_frequency, tolerance1, tolerance2)
return lpc
def resample(soundObj, target_sampling_rate, precision_ms=50):
'''Resample soundObj with the target one
Returns the updated soundObj (praat)
'''
raw_sampling_rate = call(soundObj, 'Get sampling frequency')
soundObj = call(soundObj, 'Resample', target_sampling_rate, precision_ms)
return soundObj
def relative_position(self, extremum, type, start, end):
"""
Calculate the relative position of either a maximum or minimum value within a timespan delimited by start and end timestamps
extremum: one of "maximum" and "minimum"
type: one of "pitch" and "intensity"
"""
base = extremum_at = None
if type == "pitch":
base = self.pitch_obj
elif type == "intensity":
base = self.int_obj
if type == "pitch":
extremum_at = praat.call(base, f"Get time of {extremum}", start,
end, "Hertz", "None")
elif type == "intensity":
extremum_at = praat.call(base, f"Get time of {extremum}", start,
end, "None")
time_passed = extremum_at - start
relative_pos = time_passed / (end - start)
return relative_pos
def to_formant(self,
method:str="burg",
time_step:Optional[Real]=None,
max_number_of_formants:Real=5.0,
maximum_formant:Real=5500.0,
window_length:Real=0.025,
pre_emphasis_from:Real=50.0,
number_of_std_dev:Real=1.5,
maximum_number_of_iterations:Real=5,
tolerance:Real=1.0e-6) -> pm.Formant:
"""
Parameters
----------
method: str, default "burg", can also be "sl", "keep all", "robust",
time_step: real number, optional, units in (s),
max_number_of_formants: real number, default 5.0,
maximum_formant: real number, default 5500.0, units in (Hz),
window_length: real number, default 0.025, units in (s),
pre_emphasis_from: real number, default 50.0, units in (Hz),
number_of_std_dev: real number, default 1.5,
maximum_number_of_iterations: real number, default 5,
tolerance: real number, default 1.0e-6
"""
m = method.lower()
if method == "burg":
return self.to_formant_burg(time_step, max_number_of_formants, maximum_formant, window_length, pre_emphasis_from)
elif method in ["sl", "split levinson", "split levinson (willems)"]:
return call(self, "To Formant (sl)", time_step or 0.0, max_number_of_formants, maximum_formant, window_length, pre_emphasis_from)
elif method == "keep all":
return call(self, "To Formant (keep all)", time_step or 0.0, max_number_of_formants, maximum_formant, window_length, pre_emphasis_from)
elif method == "robust":
return call(self, "To Formant (robust)", time_step or 0.0, max_number_of_formants, maximum_formant, window_length, pre_emphasis_from, number_of_std_dev, maximum_number_of_iterations, tolerance)
def extractPitch(sound, pitchFloor, pitchCeiling, unit, interpolation):
pitch = call(sound, "To Pitch", 0.0, pitchFloor, pitchCeiling)
minPitch = call(pitch, "Get minimum", 0, 0, unit, interpolation)
maxPitch = call(pitch, "Get maximum", 0, 0, unit, interpolation)
meanPitch = call(pitch, "Get mean", 0, 0, unit)
sdPitch = call(pitch, "Get standard deviation", 0, 0, unit)
return minPitch, maxPitch, meanPitch, sdPitch
def get_excursion(self, level=""):
"""
Extract the pitch excursion with normalization on either the "word" level or the intonation phrase ("ip") level
"""
if level == "word":
check_input_df(self.nuclei, ["word_start", "word_end", "f0_max"])
timestamps_filtered = self.nuclei[
(self.nuclei["word_start"].notna())
& (self.nuclei["word_end"].notna())].copy()
# Calculate 10th percentile of the pitch contour during nucleus
timestamps_filtered["f0_q10"] = [
praat.call(
self.pitch_obj,
"Get quantile",
row.word_start,
row.word_end,
0.1,
"Hertz",
) for row in timestamps_filtered.itertuples()
]
norm_df = pd.merge(self.nuclei, timestamps_filtered, how="left")
elif level == "ip":
check_input_df(self.nuclei, ["ip_start", "ip_end", "f0_max"])
nuclei_filtered = self.nuclei[(self.nuclei["ip_start"].notna()) & (
self.nuclei["ip_end"].notna())].copy()
timestamps_filtered = nuclei_filtered[["ip_start", "ip_end"
]].drop_duplicates()
# Calculate 10th percentile of the pitch contour during nucleus
timestamps_filtered["f0_q10"] = [
praat.call(
self.pitch_obj,
"Get quantile",
row.ip_start,
row.ip_end,
0.1,
"Hertz",
) for row in timestamps_filtered.itertuples()
]
norm_df = pd.merge(self.nuclei,
timestamps_filtered,
on=["ip_start", "ip_end"],
how="left")
else:
raise ValueError("Argument 'level' must be one of ['word', 'ip']")
# Calculate excursion: 12 * log2(F0_max/F0_10%)
excursions = np.array(12 *
np.log2(norm_df["f0_max"] / norm_df["f0_q10"]))
return excursions
def extractIntensity(sound, minPitch, timeStep, interpolation):
intensity = call(sound, "To Intensity", minPitch, timeStep)
minIntensity = call(intensity, "Get minimum", 0, 0, interpolation)
maxIntensity = call(intensity, "Get maximum", 0, 0, interpolation)
meanIntensity = call(intensity, "Get mean", 0, 0)
sdIntensity = call(intensity, "Get standard deviation", 0, 0)
return minIntensity, maxIntensity, meanIntensity, sdIntensity
def audio_to_textgrid(audio_no_annot_path, textgrid_path):
"""Generates .TextGrid files from audio using Praat.
:param audio_no_annot_path: path with audio not annotated.
:param textgrid_path: path with textgrid generated.
:returns: None.
"""
create_data_path(textgrid_path)
audio_files = os.listdir(audio_no_annot_path)
print("Procesing " + str(len(audio_files)) + " audio files with Praat, can take a while...")
data_mod = round(len(audio_files)/10)
index_mod = 1
for audio in audio_files:
sound = parselmouth.Sound(os.path.join(audio_no_annot_path, audio))
# Take each audio file and convert to textGrid with praat
noise_reduction = call(sound, "Remove noise", 0.0, 1.0, 0.025, 80, 10000, 40, 'Spectral subtraction') # 'silent', 'sounding'
manipulation = call(noise_reduction, "To TextGrid (silences)", 100, 0.0, -65.0, 0.8, 0.2, '', 'sounding') # 'silent', 'sounding' # 0.8, 0.2
# minimun silent and sound intervals
text_audio = audio[:-4] + '.TextGrid'
call(manipulation, "Save as text file", os.path.join(textgrid_path, text_audio))
# Show progresion
try:
if index_mod % data_mod == 0: # it means 10%, 20%, ...
print(str(int(index_mod/data_mod * 10))+"% ",end="\r")
except:
print('',end="\r")
#end try
index_mod += 1
def main(wavscp, outdir, text):
utt2text = dict()
if not path.exists(outdir):
mkdir(outdir)
if text is not None:
with open(text, 'r') as rf:
lines = rf.readlines()
for line in lines:
uttid = line.split(' ')[0]
assert uttid not in utt2text.keys(
), '[error]utterance name in text file should not be duplicated'
utt2text[uttid] = ' '.join(line.split(' ')[1:]).strip()
with open(wavscp, 'r') as rf:
lines = rf.readlines()
rule_ch_seq_space = re.compile(
r'(?<=[\u4e00-\u9fa5])( +)(?=[\u4e00-\u9fa5])')
for line in tqdm(lines):
uttid = line.split(' ')[0]
wav_path = line.split(' ')[1].strip()
wav_name = path.splitext(path.basename(wav_path))[0]
output_tgt = path.join(outdir, f'{wav_name}.TextGrid')
wav_read = praat.call('Read from file', wav_path)
tg_obj = praat.call(wav_read, 'To TextGrid', 'spk1', '')
tgt_obj = tg_obj.to_tgt()
if uttid in utt2text.keys():
text_content = rule_ch_seq_space.sub('', utt2text[uttid])
annot_text = tgt.core.Interval(tgt_obj.start_time,
tgt_obj.end_time, text_content)
tgt_obj.get_tier_by_name('spk1').add_annotation(annot_text)
tgt.io.write_to_file(tgt_obj,
output_tgt,
format='long',
encoding='utf-8')
def get_all_features(self, f0min, f0max, unit):
try:
sound = parselmouth.Sound(self.voiceID) # read the sound
pointProcess = call(sound, "To PointProcess (periodic, cc)", f0min, f0max)
Pitch = call(sound, "To Pitch", 0.0, f0min, f0max)
#Vocal Report
self.vocal_report = parselmouth.praat.call([sound, Pitch, pointProcess], "Voice report", 0, 0, 75, 600, 1.3, 1.6, 0.03, 0.45)
#Pitch
self.get_pitch_parameters(Pitch=Pitch, unit=unit)
#Harmonicity
Harmonicity = self.get_harmonicity_parameters(sound)
#Jitter
self.get_jitter_parameters(sound=sound, pointProcess=pointProcess)
#Shimmer
self.get_shimmer_parameters(sound=sound, pointProcess=pointProcess)
#Pulse
self.get_pulse_parameters()
#Voicing
self.get_voicing_parameters()
except Exception as e:
pass
self.all_vocal_parameters = self.__dict__
return self.all_vocal_parameters
def get_pitch(sound, _mean = True, _stdev= False, _range = False):
"""
Gets pitch for each audio frame of audio via parselmouth praat.
Takes mean or range of the obtained ndarray.
Parameters
----------
sound:parselmouth object
audio object
_stdev:boolean
True, if want to get standard deviation of pitch
_range:boolean
True, if want to get range of deviation of pitch
Returns
-------
float
Mean pitch of the audio sample.
Examples
--------
>>> get_stdev_energy(y)
59.78
"""
pitch = call(sound, "To Pitch", 0.0, 75, 300)
if _mean:
mean_pitch = call(pitch, "Get mean", 0, 0,'Hertz')
return mean_pitch
if _stdev:
stdev_pitch = call(pitch, "Get standard deviation", 0 ,0, "Hertz")
return stdev_pitch
if _range:
stdevPitch = call(pitch, "Get standard deviation", 0 ,0, "Hertz")
range_pitch = 4* stdevPitch
return range_pitch
def measureFormants(sound):
sound = parselmouth.Sound(sound) # read the sound
pointProcess = call(sound, "To PointProcess (periodic, cc)", 75, 500)
pitch = call(sound, "To Pitch", 0.0, 75, 500) # check pitch to set formant settings
meanF0 = call(pitch, "Get mean", 0, 0, "Hertz") # get mean pitch
if meanF0 > 150:
maxFormant = 5500
else:
maxFormant = 5000
formants = call(sound, "To Formant (burg)", 0.0025, 5, maxFormant, 0.025, 50)
numPoints = call(pointProcess, "Get number of points")
f1_list = []
f2_list = []
f3_list = []
f4_list = []
# Measure formants only at glottal pulses
for point in range(0, numPoints):
point += 1
t = call(pointProcess, "Get time from index", point)
f1 = call(formants, "Get value at time", 1, t, 'Hertz', 'Linear')
f2 = call(formants, "Get value at time", 2, t, 'Hertz', 'Linear')
f3 = call(formants, "Get value at time", 3, t, 'Hertz', 'Linear')
f4 = call(formants, "Get value at time", 4, t, 'Hertz', 'Linear')
f1_list.append(f1)
f2_list.append(f2)
f3_list.append(f3)
f4_list.append(f4)
if isinstance(f1, float) == True:
f1_list.append(f1)
if isinstance(f2, float) == True:
f2_list.append(f2)
if isinstance(f3, float) == True:
f3_list.append(f3)
if isinstance(f4, float) == True:
f4_list.append(f4)
# calculate mean formants across pulses
if len(f1_list) > 0:
f1_mean = sum(f1_list) / len(f1_list)
else:
f1_mean = "N/A"
if len(f2_list) > 0:
f2_mean = sum(f2_list) / len(f2_list)
else:
f2_mean = "N/A"
if len(f3_list) > 0:
f3_mean = sum(f3_list) / len(f3_list)
else:
f3_mean = "N/A"
if len(f4_list) > 0:
f4_mean = sum(f4_list) / len(f4_list)
else:
f4_mean = "N/A"
# calculate median formants across pulses, this is what is used in all subsequent calcualtions
# you can use mean if you want, just edit the code in the boxes below to replace median with mean
return f1_mean, f2_mean, f3_mean, f4_mean
def measurePitch(voiceID, f0min, f0max, unit):
sound = parselmouth.Sound(voiceID) # read the sound
pitch = call(sound, "To Pitch", 0.0, f0min,
f0max) #create a praat pitch object
meanfreq = call(pitch, "Get mean", 0, 0, unit) # get mean pitch
sd = call(pitch, "Get standard deviation", 0, 0,
unit) # get standard deviation
return meanfreq, sd
def extractJitterAndShimmer(sound, pitchFloor, pitchCeiling):
pitch = call(sound, "To Pitch", 0.0, pitchFloor, pitchCeiling)
pointProcess = call(pitch, "To PointProcess")
localJitter = call(pointProcess, "Get jitter (local)", 0, 0, 0.0001, 0.02,
1.3)
localShimmer = call([sound, pointProcess], "Get shimmer (local)", 0, 0,
0.0001, 0.02, 1.3, 1.6)
return localJitter, localShimmer
def get_jitter_parameters(self, sound, pointProcess):
self.jitter_local = call(pointProcess, "Get jitter (local)", 0, 0, 0.0001, 0.02, 1.3)
self.jitter_absolute_local = call(pointProcess, "Get jitter (local, absolute)", 0, 0, 0.0001, 0.02, 1.3)
self.jitter_rap = call(pointProcess, "Get jitter (rap)", 0, 0, 0.0001, 0.02, 1.3)
self.jitter_ppq5 = call(pointProcess, "Get jitter (ppq5)", 0, 0, 0.0001, 0.02, 1.3)
self.jitter_ddp = call(pointProcess, "Get jitter (ddp)", 0, 0, 0.0001, 0.02, 1.3)
return None
def manipulateFormants(wav_file, gender, factor):
sound = parselmouth.Sound(wav_file)
if gender == "female":
manipulated_sound = call(sound, "Change gender", 60, 300, factor, 0, 1,
1)
elif gender == "male":
manipulated_sound = call(sound, "Change gender", 100, 500, factor, 0,
1, 1)
return manipulated_sound
def validate(intensity_obj, peak_cands):
""" This function validates the n potential peaks (i.e. potential syllable nuclei) that were found by checking whether they are:
- followed by a min. 2dB dip (first peak)
- surrounded by min. 2dB dip (second to penultimate peak) -- DISABLED
- min. dB dip on any side (second to penultimate peak)
- preceded by a min. 2dB dip (last peak)
"""
valid_peaks = []
for i in range(len(peak_cands) - 1):
peak = peak_cands[i]
if i == 0:
next_peak = peak_cands[i + 1]
next_intensity_dip = praat.call(intensity_obj, "Get minimum",
peak[0], next_peak[0], "None")
intensity_diff = abs(peak[1] - next_intensity_dip)
if intensity_diff > MIN_DIP_BETW_PEAKS:
valid_peaks.append(peak)
elif 0 < i < len(peak_cands) - 1:
next_peak = peak_cands[i + 1]
next_intensity_dip = praat.call(intensity_obj, "Get minimum",
peak[0], next_peak[0], "None")
intensity_diff = abs(peak[1] - next_intensity_dip)
if intensity_diff > MIN_DIP_BETW_PEAKS:
# DISABLED: Possibility to require a dip before the nucleus as well.
"""prev_peak = peak_cands[i - 1]
prev_intensity_dip = praat.call(
intensity_obj, "Get minimum", prev_peak[0], peak[0], "None"
)
intensity_diff = abs(peak[1] - prev_intensity_dip)
if intensity_diff > MIN_DIP_BETW_PEAKS:"""
valid_peaks.append(peak)
else:
prev_peak = peak_cands[i - 1]
prev_intensity_dip = praat.call(intensity_obj, "Get minimum",
prev_peak[0], peak[0], "None")
intensity_diff = abs(peak[1] - prev_intensity_dip)
if intensity_diff > MIN_DIP_BETW_PEAKS:
valid_peaks.append(peak)
else:
prev_peak = peak_cands[i - 1]
prev_intensity_dip = praat.call(intensity_obj, "Get minimum",
prev_peak[0], peak[0], "None")
intensity_diff = abs(peak[1] - prev_intensity_dip)
if intensity_diff > MIN_DIP_BETW_PEAKS:
valid_peaks.append(peak)
return valid_peaks
def get_formants(phones):
wav_file = phones.loc[0, "wav"]
snd = parselmouth.Sound(wav_file)
formants = call(snd, "To Formant (burg)", 0.0025, 5, 5000, 0.025, 50)
for i in range(1, 5):
phones["f{}".format(i)] = phones.apply(lambda r: np.nan if r["phone"] not in TimitData.VOWELS
else np.array([call(formants, "Get value at time", i, t, 'Hertz', 'Linear') for t in
TimitData.get_formant_times(r)]),
axis=1)
return phones
def get_avg_pitch(infile):
sound = parselmouth.Sound(infile)
Audio(data=sound.values, rate=sound.sampling_frequency)
manipulation = call(sound, "To Manipulation", 0.001, 75, 600)
pitch_tier = call(manipulation, "Extract pitch tier")
pitch = sound.to_pitch()
pitch_values = pitch.selected_array['frequency']
# remove values lower than 65Hz (that's about the lowest freq for male voice)
pitch_values = list(filter(lambda bigval: bigval >= 65, pitch_values))
return np.mean(pitch_values)
def save_pitch_and_pulse(sound):
manipulation = call(sound, "To Manipulation", 0.01, 75, 600)
# Save pitch data
pitch_tier = call(manipulation, "Extract pitch tier")
pitch_tier_loc = '{}/vocals.PitchTier'.format(output_folder)
pitch_tier.save_as_text_file(pitch_tier_loc)
# Save pulse data
pulse = call(manipulation, "Extract pulses")
pulse_loc = '{}/vocals.Pulse'.format(output_folder)
pulse.save_as_text_file(pulse_loc)
def measure_pitch(
voice,
floor=50,
ceiling=500,
method="ac",
time_step=0,
max_number_of_candidates=15,
silence_threshold=0.03,
voicing_threshold=0.45,
octave_cost=0.01,
octave_jump_cost=0.35,
voiced_unvoiced_cost=0.14,
unit="Hertz",
very_accurate="no",
):
#floor, ceiling = pitch_bounds(voice)
"""
Args:
voice:
floor:
ceiling:
method:
time_step:
max_number_of_candidates:
silence_threshold:
voicing_threshold:
octave_cost:
octave_jump_cost:
voiced_unvoiced_cost:
unit:
very_accurate:
"""
pitch: object = call(
voice,
method,
time_step,
floor,
max_number_of_candidates,
very_accurate,
silence_threshold,
voicing_threshold,
octave_cost,
octave_jump_cost,
voiced_unvoiced_cost,
ceiling,
)
mean_f0: float = call(pitch, "Get mean", 0, 0, unit)
stdev_f0: float = call(pitch, "Get standard deviation", 0, 0,
unit) # get standard deviation
min_f0: float = call(pitch, "Get minimum", 0, 0, unit, "Parabolic")
max_f0: float = call(pitch, "Get maximum", 0, 0, unit, "Parabolic")
return pitch, mean_f0, stdev_f0, min_f0, max_f0
def get_pitch_parameters(self, Pitch, unit):
self.pitch_mean = call(Pitch, "Get mean", 0, 0, unit) # get mean pitch
self.pitch_median = float(re.findall("Median pitch: ([0-9]*\.[0-9]*)",
self.vocal_report)[0])
self.pitch_std_dev = call(Pitch, "Get standard deviation", 0, 0, unit) # get standard deviation
self.pitch_minimum = float(re.findall("Minimum pitch: ([0-9]*\.[0-9]*)",
self.vocal_report)[0])
self.pitch_maximum = float(re.findall("Maximum pitch: ([0-9]*\.[0-9]*)",
self.vocal_report)[0])
return None
def process(self):
sound = self.args["voice"]
formant_factor = self.args["formant_factor"]
pitch_factor = self.args["pitch_factor"]
duration = sound.get_total_duration()
file_path = self.args["file_path"]
pitch_range_factor = self.args["pitch_range_factor"]
duration_factor = 1
pitch_range_factor = 1
f0min, f0max = self.pitch_bounds(sound)
pitch = sound.to_pitch()
print(f0min, f0max)
median_pitch = call(pitch, "Get quantile", sound.xmin, sound.xmax, 0.5,
"Hertz")
print(
f"mean pitch {call(pitch, 'Get mean', sound.xmin, sound.xmax, 'Hertz' )}"
)
if formant_factor > 1:
formant_factor = 1 / formant_factor
if pitch_factor > 1:
pitch_factor = 1 / pitch_factor
print(median_pitch)
print(pitch_factor)
new_pitch_median = pitch_factor * median_pitch
print(new_pitch_median)
output_file_name = file_path.split("/")[-1].split(".wav")[0]
output_file_name = (
f"{output_file_name}_raise_pitch_and_formants_{pitch_factor}_{formant_factor}"
)
manipulated_sound = call(
sound,
"Change gender",
f0min,
f0max,
formant_factor,
new_pitch_median,
pitch_range_factor,
duration_factor,
)
if self.args["normalize amplitude"]:
manipulated_sound.scale_intensity(70)
manipulated_sound.name = output_file_name
return {"voice": manipulated_sound}
def get_silence_threshold(sound, lower_quantile):
""" Calculates silence threshold per sound interval for chunking.
:param sound: A parselmouth.praat Sound object
:param lower_quantile: A quantile value (0-1; e.g., 0.5 = median)
:return sil_threshold: Threshold value to be used for 'To TextGrid (silences)'
"""
soundint = sound.to_intensity()
max_intensity = call(soundint, 'Get quantile', 0.0, 0.0, 1)
sil_intensity = call(soundint, 'Get quantile', 0.0, 0.0, lower_quantile)
return sil_intensity - max_intensity
def chunk_sound (sound, sil_duration, threshold_quantile):
sil_threshold = get_silence_threshold(sound, threshold_quantile)
textgrid = detect_silences(sound, sil_threshold, sil_duration)
n_ints = call(textgrid, 'Count intervals where',
1, 'is equal to', 'speech')
extracted_sounds = call([sound, textgrid],
'Extract intervals where',
1, True, 'is equal to', 'speech')
return textgrid, extracted_sounds, n_ints
def measure_jitter(self):
self.point_process: object = call(self.sound,
"To PointProcess (periodic, cc)", 60,
600)
self.local_jitter_teva: float = call(
self.point_process,
"Get jitter (local)",
self.start_time,
self.end_time,
self.shortest_period,
self.longest_period,
self.maximum_period_factor,
)
def measure_mvd(self):
# Maximum Voicing Duration (MVD)
textgrid = call(self.point_process, "To TextGrid (vuv)", 0.2, 0.1)
maximum_voicing_durations = []
number_of_intervals = call(textgrid, "Get number of intervals", 1)
for interval, number in enumerate(range(number_of_intervals), 1):
label = call(textgrid, "Get label of interval", 1, interval)
if "v" in label.lower():
start = call(textgrid, "Get start point", 1, interval)
end = call(textgrid, "Get end point", 1, interval)
maximum_voicing_duration = end - start
maximum_voicing_durations.append(maximum_voicing_duration)
self.mvd = max(maximum_voicing_durations)
