def overlap_and_add(signal, frame_step):
"""Reconstructs a signal from a framed representation.
Adds potentially overlapping frames of a signal with shape
`[..., frames, frame_length]`, offsetting subsequent frames by `frame_step`.
The resulting tensor has shape `[..., output_size]` where
output_size = (frames - 1) * frame_step + frame_length
Args:
signal: A [..., frames, frame_length] Tensor. All dimensions may be unknown, and rank must be at least 2.
frame_step: An integer denoting overlap offsets. Must be less than or equal to frame_length.
Returns:
A Tensor with shape [..., output_size] containing the overlap-added frames of signal's inner-most two dimensions.
output_size = (frames - 1) * frame_step + frame_length
Based on https://github.com/tensorflow/tensorflow/blob/r1.12/tensorflow/contrib/signal/python/ops/reconstruction_ops.py
"""
outer_dimensions = signal.size()[:-2]
frames, frame_length = signal.size()[-2:]
subframe_length = math.gcd(frame_length, frame_step) # gcd=Greatest Common Divisor
subframe_step = frame_step // subframe_length
subframes_per_frame = frame_length // subframe_length
output_size = frame_step * (frames - 1) + frame_length
output_subframes = output_size // subframe_length
subframe_signal = signal.view(*outer_dimensions, -1, subframe_length)
frame = torch.arange(0, output_subframes).unfold(0, subframes_per_frame, subframe_step)
frame = signal.new_tensor(frame).long() # signal may in GPU or CPU
frame = frame.contiguous().view(-1)
result = signal.new_zeros(*outer_dimensions, output_subframes, subframe_length)
result.index_add_(-2, frame, subframe_signal)
result = result.view(*outer_dimensions, -1)
return result
def percentile(signal, p):
"""Calculate percentile of signal
Args:
signal (np.array/th.tensor): Signal to normalize
p (int): [0-100]. Percentile to find
Returns:
int: Percentile signal value
"""
k = 1 + round(0.01 * float(p) * (signal.numel() - 1))
return signal.view(-1).kthvalue(k).values.item()