File size: 5,029 Bytes
9c6594c |
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 |
# Copyright The Lightning team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import Any
import numpy as np
import torch
from torch import Tensor
from torchmetrics.utilities.checks import _check_same_shape
from torchmetrics.utilities.imports import _MULTIPROCESSING_AVAILABLE, _PESQ_AVAILABLE
__doctest_requires__ = {("perceptual_evaluation_speech_quality",): ["pesq"]}
def perceptual_evaluation_speech_quality(
preds: Tensor,
target: Tensor,
fs: int,
mode: str,
keep_same_device: bool = False,
n_processes: int = 1,
) -> Tensor:
r"""Calculate `Perceptual Evaluation of Speech Quality`_ (PESQ).
It's a recognized industry standard for audio quality that takes into considerations characteristics such as: audio
sharpness, call volume, background noise, clipping, audio interference etc. PESQ returns a score between -0.5 and
4.5 with the higher scores indicating a better quality.
This metric is a wrapper for the `pesq package`_. Note that input will be moved to `cpu` to perform the metric
calculation.
.. hint::
Usingsing this metrics requires you to have ``pesq`` install. Either install as ``pip install
torchmetrics[audio]`` or ``pip install pesq``. Note that ``pesq`` will compile with your currently
installed version of numpy, meaning that if you upgrade numpy at some point in the future you will
most likely have to reinstall ``pesq``.
Args:
preds: float tensor with shape ``(...,time)``
target: float tensor with shape ``(...,time)``
fs: sampling frequency, should be 16000 or 8000 (Hz)
mode: ``'wb'`` (wide-band) or ``'nb'`` (narrow-band)
keep_same_device: whether to move the pesq value to the device of preds
n_processes: integer specifying the number of processes to run in parallel for the metric calculation.
Only applies to batches of data and if ``multiprocessing`` package is installed.
Returns:
Float tensor with shape ``(...,)`` of PESQ values per sample
Raises:
ModuleNotFoundError:
If ``pesq`` package is not installed
ValueError:
If ``fs`` is not either ``8000`` or ``16000``
ValueError:
If ``mode`` is not either ``"wb"`` or ``"nb"``
RuntimeError:
If ``preds`` and ``target`` do not have the same shape
Example:
>>> from torch import randn
>>> from torchmetrics.functional.audio.pesq import perceptual_evaluation_speech_quality
>>> preds = randn(8000)
>>> target = randn(8000)
>>> perceptual_evaluation_speech_quality(preds, target, 8000, 'nb')
tensor(2.2885)
>>> perceptual_evaluation_speech_quality(preds, target, 16000, 'wb')
tensor(1.6805)
"""
if not _PESQ_AVAILABLE:
raise ModuleNotFoundError(
"PESQ metric requires that pesq is installed."
" Either install as `pip install torchmetrics[audio]` or `pip install pesq`."
)
import pesq as pesq_backend
def _issubtype_number(x: Any) -> bool:
return np.issubdtype(type(x), np.number)
_filter_error_msg = np.vectorize(_issubtype_number)
if fs not in (8000, 16000):
raise ValueError(f"Expected argument `fs` to either be 8000 or 16000 but got {fs}")
if mode not in ("wb", "nb"):
raise ValueError(f"Expected argument `mode` to either be 'wb' or 'nb' but got {mode}")
_check_same_shape(preds, target)
if preds.ndim == 1:
pesq_val_np = pesq_backend.pesq(fs, target.detach().cpu().numpy(), preds.detach().cpu().numpy(), mode)
pesq_val = torch.tensor(pesq_val_np)
else:
preds_np = preds.reshape(-1, preds.shape[-1]).detach().cpu().numpy()
target_np = target.reshape(-1, preds.shape[-1]).detach().cpu().numpy()
if _MULTIPROCESSING_AVAILABLE and n_processes != 1:
pesq_val_np = pesq_backend.pesq_batch(fs, target_np, preds_np, mode, n_processor=n_processes)
pesq_val_np = np.array(pesq_val_np)
else:
pesq_val_np = np.empty(shape=(preds_np.shape[0]))
for b in range(preds_np.shape[0]):
pesq_val_np[b] = pesq_backend.pesq(fs, target_np[b, :], preds_np[b, :], mode)
pesq_val = torch.from_numpy(pesq_val_np[_filter_error_msg(pesq_val_np)].astype(np.float32))
pesq_val = pesq_val.reshape(len(pesq_val))
if keep_same_device:
return pesq_val.to(preds.device)
return pesq_val
|