def run_vad(self):
"""
Determine which frames contain speech and nonspeech,
and store the resulting boolean mask internally.
"""
def _compute_runs(array):
"""
Compute runs as a list of arrays,
each containing the indices of a contiguous run.
:param array: the data array
:type array: :class:`numpy.ndarray` (1D)
:rtype: list of :class:`numpy.ndarray` (1D)
"""
if len(array) < 1:
return []
return numpy.split(array,
numpy.where(numpy.diff(array) != 1)[0] + 1)
self.log(u"Creating VAD object")
vad = VAD(rconf=self.rconf, logger=self.logger)
self.log(u"Running VAD...")
self.__mfcc_mask = vad.run_vad(self.__mfcc[0])
self.__mfcc_mask_map = (numpy.where(self.__mfcc_mask))[0]
self.log(u"Running VAD... done")
self.log(u"Storing speech and nonspeech intervals...")
# where( == True) already computed, reusing
# COMMENTED runs = _compute_runs((numpy.where(self.__mfcc_mask))[0])
runs = _compute_runs(self.__mfcc_mask_map)
self.__speech_intervals = [(r[0], r[-1]) for r in runs]
# where( == False) not already computed, computing now
runs = _compute_runs((numpy.where(~self.__mfcc_mask))[0])
self.__nonspeech_intervals = [(r[0], r[-1]) for r in runs]
self.log(u"Storing speech and nonspeech intervals... done")
def run_vad(self):
"""
Determine which frames contain speech and nonspeech,
and store the resulting boolean mask internally.
"""
def _compute_runs(array):
"""
Compute runs as a list of arrays,
each containing the indices of a contiguous run.
:param array: the data array
:type array: :class:`numpy.ndarray` (1D)
:rtype: list of :class:`numpy.ndarray` (1D)
"""
if len(array) < 1:
return []
return numpy.split(array, numpy.where(numpy.diff(array) != 1)[0] + 1)
self.log(u"Creating VAD object")
vad = VAD(rconf=self.rconf, logger=self.logger)
self.log(u"Running VAD...")
self.__mfcc_mask = vad.run_vad(self.__mfcc[0])
self.__mfcc_mask_map = (numpy.where(self.__mfcc_mask))[0]
self.log(u"Running VAD... done")
self.log(u"Storing speech and nonspeech intervals...")
# where( == True) already computed, reusing
#runs = _compute_runs((numpy.where(self.__mfcc_mask))[0])
runs = _compute_runs(self.__mfcc_mask_map)
self.__speech_intervals = [(r[0], r[-1]) for r in runs]
# where( == False) not already computed, computing now
runs = _compute_runs((numpy.where(~self.__mfcc_mask))[0])
self.__nonspeech_intervals = [(r[0], r[-1]) for r in runs]
self.log(u"Storing speech and nonspeech intervals... done")
def run_vad(
self,
log_energy_threshold=None,
min_nonspeech_length=None,
extend_before=None,
extend_after=None
):
"""
Determine which frames contain speech and nonspeech,
and store the resulting boolean mask internally.
The four parameters might be ``None``:
in this case, the corresponding RuntimeConfiguration values
are applied.
:param float log_energy_threshold: the minimum log energy threshold to consider a frame as speech
:param int min_nonspeech_length: the minimum length, in frames, of a nonspeech interval
:param int extend_before: extend each speech interval by this number of frames to the left (before)
:param int extend_after: extend each speech interval by this number of frames to the right (after)
"""
def _compute_runs(array):
"""
Compute runs as a list of arrays,
each containing the indices of a contiguous run.
:param array: the data array
:type array: :class:`numpy.ndarray` (1D)
:rtype: list of :class:`numpy.ndarray` (1D)
"""
if len(array) < 1:
return []
return numpy.split(array, numpy.where(numpy.diff(array) != 1)[0] + 1)
self.log(u"Creating VAD object")
vad = VAD(rconf=self.rconf, logger=self.logger)
self.log(u"Running VAD...")
self.__mfcc_mask = vad.run_vad(
wave_energy=self.__mfcc[0],
log_energy_threshold=log_energy_threshold,
min_nonspeech_length=min_nonspeech_length,
extend_before=extend_before,
extend_after=extend_after
)
self.__mfcc_mask_map = (numpy.where(self.__mfcc_mask))[0]
self.log(u"Running VAD... done")
self.log(u"Storing speech and nonspeech intervals...")
# where( == True) already computed, reusing
# COMMENTED runs = _compute_runs((numpy.where(self.__mfcc_mask))[0])
runs = _compute_runs(self.__mfcc_mask_map)
self.__speech_intervals = [(r[0], r[-1]) for r in runs]
# where( == False) not already computed, computing now
runs = _compute_runs((numpy.where(~self.__mfcc_mask))[0])
self.__nonspeech_intervals = [(r[0], r[-1]) for r in runs]
self.log(u"Storing speech and nonspeech intervals... done")
def run_vad(self,
log_energy_threshold=None,
min_nonspeech_length=None,
extend_before=None,
extend_after=None):
"""
Determine which frames contain speech and nonspeech,
and store the resulting boolean mask internally.
The four parameters might be ``None``:
in this case, the corresponding RuntimeConfiguration values
are applied.
:param float log_energy_threshold: the minimum log energy threshold to consider a frame as speech
:param int min_nonspeech_length: the minimum length, in frames, of a nonspeech interval
:param int extend_before: extend each speech interval by this number of frames to the left (before)
:param int extend_after: extend each speech interval by this number of frames to the right (after)
"""
def _compute_runs(array):
"""
Compute runs as a list of arrays,
each containing the indices of a contiguous run.
:param array: the data array
:type array: :class:`numpy.ndarray` (1D)
:rtype: list of :class:`numpy.ndarray` (1D)
"""
if len(array) < 1:
return []
return numpy.split(array,
numpy.where(numpy.diff(array) != 1)[0] + 1)
self.log(u"Creating VAD object")
vad = VAD(rconf=self.rconf, logger=self.logger)
self.log(u"Running VAD...")
self.__mfcc_mask = vad.run_vad(
wave_energy=self.__mfcc[0],
log_energy_threshold=log_energy_threshold,
min_nonspeech_length=min_nonspeech_length,
extend_before=extend_before,
extend_after=extend_after)
self.__mfcc_mask_map = (numpy.where(self.__mfcc_mask))[0]
self.log(u"Running VAD... done")
self.log(u"Storing speech and nonspeech intervals...")
# where( == True) already computed, reusing
# COMMENTED runs = _compute_runs((numpy.where(self.__mfcc_mask))[0])
runs = _compute_runs(self.__mfcc_mask_map)
self.__speech_intervals = [(r[0], r[-1]) for r in runs]
# where( == False) not already computed, computing now
runs = _compute_runs((numpy.where(~self.__mfcc_mask))[0])
self.__nonspeech_intervals = [(r[0], r[-1]) for r in runs]
self.log(u"Storing speech and nonspeech intervals... done")
