# Copyright 2023 Yifeng Yu
# Apache 2.0 (http://www.apache.org/licenses/LICENSE-2.0)
"""Function to get random segments."""
import torch.nn.functional as F
[docs]def expand_f0(f0_frame, hop_length, method="interpolation"):
"""Expand f0 to output wave length.
Args:
f0_frame (Tensor): Input tensor (B, 1, frame_len).
hop_length (Tensor): Hop length.
method (str): Method to expand f0. Choose either 'interpolation' or 'repeat'.
Returns:
Tensor: Output tensor (B, 1, wav_len).
"""
frame_length = f0_frame.size(2)
signal_length = frame_length * hop_length
if method == "interpolation":
f0_sample = F.interpolate(
f0_frame, size=signal_length, mode="linear", align_corners=False
)
elif method == "repeat":
f0_sample = f0_frame.repeat_interleave(hop_length, dim=2)[:signal_length]
else:
raise ValueError("Invalid method. Choose either 'interpolation' or 'repeat'.")
f0_sample = f0_sample.squeeze()[
:signal_length
] # Remove extra dimensions and trim to signal_length
return f0_sample