File size: 4,231 Bytes
51e2f90 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 |
from typing import Any, Callable
import numpy as np
import torch
import torchmetrics as tm
from torch._C import _LinAlgError
from torchmetrics import functional as tmF
class SafeSignalDistortionRatio(tm.SignalDistortionRatio):
def __init__(self, **kwargs) -> None:
super().__init__(**kwargs)
def update(self, *args, **kwargs) -> Any:
try:
super().update(*args, **kwargs)
except:
pass
def compute(self) -> Any:
if self.total == 0:
return torch.tensor(torch.nan)
return super().compute()
class BaseChunkMedianSignalRatio(tm.Metric):
def __init__(
self,
func: Callable,
window_size: int,
hop_size: int = None,
zero_mean: bool = False,
) -> None:
super().__init__()
# self.zero_mean = zero_mean
self.func = func
self.window_size = window_size
if hop_size is None:
hop_size = window_size
self.hop_size = hop_size
self.add_state(
"sum_snr",
default=torch.tensor(0.0),
dist_reduce_fx="sum"
)
self.add_state("total", default=torch.tensor(0), dist_reduce_fx="sum")
def update(self, preds: torch.Tensor, target: torch.Tensor) -> None:
n_samples = target.shape[-1]
n_chunks = int(
np.ceil((n_samples - self.window_size) / self.hop_size) + 1
)
snr_chunk = []
for i in range(n_chunks):
start = i * self.hop_size
if n_samples - start < self.window_size:
continue
end = start + self.window_size
try:
chunk_snr = self.func(
preds[..., start:end],
target[..., start:end]
)
# print(preds.shape, chunk_snr.shape)
if torch.all(torch.isfinite(chunk_snr)):
snr_chunk.append(chunk_snr)
except _LinAlgError:
pass
snr_chunk = torch.stack(snr_chunk, dim=-1)
snr_batch, _ = torch.nanmedian(snr_chunk, dim=-1)
self.sum_snr += snr_batch.sum()
self.total += snr_batch.numel()
def compute(self) -> Any:
return self.sum_snr / self.total
class ChunkMedianSignalNoiseRatio(BaseChunkMedianSignalRatio):
def __init__(
self,
window_size: int,
hop_size: int = None,
zero_mean: bool = False
) -> None:
super().__init__(
func=tmF.signal_noise_ratio,
window_size=window_size,
hop_size=hop_size,
zero_mean=zero_mean,
)
class ChunkMedianScaleInvariantSignalNoiseRatio(BaseChunkMedianSignalRatio):
def __init__(
self,
window_size: int,
hop_size: int = None,
zero_mean: bool = False
) -> None:
super().__init__(
func=tmF.scale_invariant_signal_noise_ratio,
window_size=window_size,
hop_size=hop_size,
zero_mean=zero_mean,
)
class ChunkMedianSignalDistortionRatio(BaseChunkMedianSignalRatio):
def __init__(
self,
window_size: int,
hop_size: int = None,
zero_mean: bool = False
) -> None:
super().__init__(
func=tmF.signal_distortion_ratio,
window_size=window_size,
hop_size=hop_size,
zero_mean=zero_mean,
)
class ChunkMedianScaleInvariantSignalDistortionRatio(
BaseChunkMedianSignalRatio
):
def __init__(
self,
window_size: int,
hop_size: int = None,
zero_mean: bool = False
) -> None:
super().__init__(
func=tmF.scale_invariant_signal_distortion_ratio,
window_size=window_size,
hop_size=hop_size,
zero_mean=zero_mean,
)
|