def DetectHardClipping(cls, signal, threshold=2):
"""Detects hard clipping.
Hard clipping is simply detected by counting samples that touch either the
lower or upper bound too many times in a row (according to |threshold|).
The presence of a single sequence of samples meeting such property is enough
to label the signal as hard clipped.
Args:
signal: AudioSegment instance.
threshold: minimum number of samples at full-scale in a row.
Returns:
True if hard clipping is detect, False otherwise.
"""
if signal.channels != 1:
raise NotImplementedError('mutliple-channel clipping not implemented')
if signal.sample_width != 2: # Note that signal.sample_width is in bytes.
raise exceptions.SignalProcessingException(
'hard-clipping detection only supported for 16 bit samples')
# Get raw samples, check type, cast.
samples = signal.get_array_of_samples()
if samples.typecode != 'h':
raise exceptions.SignalProcessingException(
'hard-clipping detection only supported for 16 bit samples')
samples = np.array(signal.get_array_of_samples(), np.int16)
# Detect adjacent clipped samples.
samples_type_info = np.iinfo(samples.dtype)
mask_min = samples == samples_type_info.min
mask_max = samples == samples_type_info.max
def HasLongSequence(vector, min_legth=threshold):
"""Returns True if there are one or more long sequences of True flags."""
seq_length = 0
for b in vector:
seq_length = seq_length + 1 if b else 0
if seq_length >= min_legth:
return True
return False
return HasLongSequence(mask_min) or HasLongSequence(mask_max)
def count_samples(cls, signal):
"""
Number of samples per channel.
Args:
signal: AudioSegment instance.
"""
number_of_samples = len(signal.get_array_of_samples())
assert signal.channels > 0
assert number_of_samples % signal.channels == 0
return number_of_samples / signal.channels
def CountSamples(cls, signal):
"""Number of samples per channel.
Args:
signal: AudioSegment instance.
Returns:
An integer.
"""
number_of_samples = len(signal.get_array_of_samples())
assert signal.channels > 0
assert number_of_samples % signal.channels == 0
return number_of_samples / signal.channels
def ApplyImpulseResponse(cls, signal, impulse_response):
"""Applies an impulse response to a signal.
Args:
signal: AudioSegment instance.
impulse_response: list or numpy vector of float values.
Returns:
AudioSegment instance.
"""
# Get samples.
assert signal.channels == 1, (
'multiple-channel recordings not supported')
samples = signal.get_array_of_samples()
# Convolve.
logging.info(
'applying %d order impulse response to a signal lasting %d ms',
len(impulse_response), len(signal))
convolved_samples = scipy.signal.fftconvolve(in1=samples,
in2=impulse_response,
mode='full').astype(
np.int16)
logging.info('convolution computed')
# Cast.
convolved_samples = array.array(signal.array_type, convolved_samples)
# Verify.
logging.debug('signal length: %d samples', len(samples))
logging.debug('convolved signal length: %d samples',
len(convolved_samples))
assert len(convolved_samples) > len(samples)
# Generate convolved signal AudioSegment instance.
convolved_signal = pydub.AudioSegment(data=convolved_samples,
metadata={
'sample_width':
signal.sample_width,
'frame_rate':
signal.frame_rate,
'frame_width':
signal.frame_width,
'channels': signal.channels,
})
assert len(convolved_signal) > len(signal)
return convolved_signal
def Copy(cls, signal):
"""Makes a copy os a signal.
Args:
signal: AudioSegment instance.
Returns:
An AudioSegment instance.
"""
return pydub.AudioSegment(data=signal.get_array_of_samples(),
metadata={
'sample_width': signal.sample_width,
'frame_rate': signal.frame_rate,
'frame_width': signal.frame_width,
'channels': signal.channels,
})
def Copy(cls, signal):
"""Makes a copy os a signal.
Args:
signal: AudioSegment instance.
Returns:
An AudioSegment instance.
"""
return pydub.AudioSegment(
data=signal.get_array_of_samples(),
metadata={
'sample_width': signal.sample_width,
'frame_rate': signal.frame_rate,
'frame_width': signal.frame_width,
'channels': signal.channels,
})
def ApplyImpulseResponse(cls, signal, impulse_response):
"""Applies an impulse response to a signal.
Args:
signal: AudioSegment instance.
impulse_response: list or numpy vector of float values.
Returns:
AudioSegment instance.
"""
# Get samples.
assert signal.channels == 1, (
'multiple-channel recordings not supported')
samples = signal.get_array_of_samples()
# Convolve.
logging.info('applying %d order impulse response to a signal lasting %d ms',
len(impulse_response), len(signal))
convolved_samples = scipy.signal.fftconvolve(
in1=samples,
in2=impulse_response,
mode='full').astype(np.int16)
logging.info('convolution computed')
# Cast.
convolved_samples = array.array(signal.array_type, convolved_samples)
# Verify.
logging.debug('signal length: %d samples', len(samples))
logging.debug('convolved signal length: %d samples', len(convolved_samples))
assert len(convolved_samples) > len(samples)
# Generate convolved signal AudioSegment instance.
convolved_signal = pydub.AudioSegment(
data=convolved_samples,
metadata={
'sample_width': signal.sample_width,
'frame_rate': signal.frame_rate,
'frame_width': signal.frame_width,
'channels': signal.channels,
})
assert len(convolved_signal) > len(signal)
return convolved_signal
def AudioSegmentToRawData(cls, signal):
samples = signal.get_array_of_samples()
if samples.typecode != 'h':
raise exceptions.SignalProcessingException(
'Unsupported samples type')
return np.array(signal.get_array_of_samples(), np.int16)
def AudioSegmentToRawData(cls, signal):
samples = signal.get_array_of_samples()
if samples.typecode != 'h':
raise exceptions.SignalProcessingException('Unsupported samples type')
return np.array(signal.get_array_of_samples(), np.int16)
