def no_overlap(D):
if D == 1:
x_local = x[:, 0]
else:
x_local = x[:, :D]
hop = block_size
# analysis
X = analysis(x_local, L=block_size, hop=hop)
# synthesis
x_r = synthesis(X, L=block_size, hop=hop)
return pra.dB(np.max(np.abs(x_local - x_r)))
def half_overlap(D):
if D == 1:
x_local = x[:, 0]
else:
x_local = x[:, :D]
hop = block_size//2
# analysis
analysis_win = pra.hann(block_size)
X = analysis(x_local, L=block_size, hop=hop, win=analysis_win)
# synthesis
x_r = synthesis(X, L=block_size, hop=hop)
return pra.dB(np.max(np.abs(x_local[:-hop, ] - x_r[hop:, ])))
def hop_one_sample(D):
if D == 1:
x_local = x[:, 0]
else:
x_local = x[:, :D]
hop = 1
# analysis
analysis_win = pra.hann(block_size)
X = analysis(x_local, L=block_size, hop=hop, win=analysis_win)
# synthesis
synthesis_win = pra.transform.compute_synthesis_window(analysis_win, hop)
x_r = synthesis(X, L=block_size, hop=hop, win=synthesis_win)
return pra.dB(
np.max(np.abs(x_local[:-block_size + hop, ] -
x_r[block_size - hop:, ])))
def append_one_sample(D):
hop = block_size // 2
n_samples = x.shape[0]
n_frames = n_samples // hop
x_local = x[:n_frames * hop - 1, :]
if D == 1:
x_local = x_local[:, 0]
else:
x_local = x_local[:, :D]
# analysis
analysis_win = pra.hann(block_size)
X = analysis(x_local, L=block_size, hop=hop, win=analysis_win)
# synthesis
x_r = synthesis(X, L=block_size, hop=hop)
return pra.dB(
np.max(
np.abs(x_local[:-block_size + hop, ] -
x_r[block_size - hop:-1, ])))
