mirror of
https://github.com/pyannote/pyannote-audio.git
synced 2026-11-20 13:07:51 +09:00
* feat(cli): add --subfolder option to benchmark command (#2063) * fix(utils): fix `SpeakerDiarizationMixin.optimal_mapping` when `reference` is a mapping * fix(task): fix `balance` option crashing training with `KeyError: 'metadata'` (#2067) * fix(io): fix `Audio.crop` raising when an in-memory waveform is cropped up to its very end (#2068) * fix(model): make Model.default_activation return identity for unsupported tasks * fix: compute sample duration from preloaded waveform * fix(pipeline): omit unset runtime options from model references (#2080) Co-authored-by: abeeranajam31 <128603901+abeeranajam31@users.noreply.github.com> Co-authored-by: antoinelaurent <antoinelaurent@me.com> Co-authored-by: Kayvan Zahiri <123409429+Kayvan-Zahiri@users.noreply.github.com> Co-authored-by: Mohammad Hijjawi <51020382+MohammadHijjawi97@users.noreply.github.com>
142 lines
5.3 KiB
Python
142 lines
5.3 KiB
Python
# The MIT License (MIT)
|
|
|
|
# Copyright (c) 2022-2025 CNRS
|
|
# Copyright (c) 2025- pyannoteAI
|
|
|
|
# Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
# of this software and associated documentation files (the "Software"), to deal
|
|
# in the Software without restriction, including without limitation the rights
|
|
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
# copies of the Software, and to permit persons to whom the Software is
|
|
# furnished to do so, subject to the following conditions:
|
|
|
|
# The above copyright notice and this permission notice shall be included in
|
|
# all copies or substantial portions of the Software.
|
|
|
|
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
# SOFTWARE.
|
|
|
|
# AUTHORS
|
|
# Hervé BREDIN - http://herve.niderb.fr
|
|
|
|
from functools import reduce
|
|
|
|
from pyannote.core import Annotation, Segment
|
|
from pyannote.database import ProtocolFile
|
|
|
|
from pyannote.audio.core.io import Audio, get_audio_metadata
|
|
|
|
|
|
class LowerTemporalResolution:
|
|
"""Artificially degrade temporal resolution of reference annotation
|
|
|
|
Parameters
|
|
----------
|
|
resolution : float, optional
|
|
Target temporal resolution, in seconds. Defaults to 0.1 (100ms).
|
|
"""
|
|
|
|
preprocessed_key = "annotation"
|
|
|
|
def __init__(self, resolution: float = 0.1):
|
|
super().__init__()
|
|
self.resolution = resolution
|
|
|
|
def __call__(self, current_file: ProtocolFile) -> Annotation:
|
|
annotation = current_file["annotation"]
|
|
new_annotation = annotation.empty()
|
|
|
|
for new_track, (segment, _, label) in enumerate(
|
|
annotation.itertracks(yield_label=True)
|
|
):
|
|
new_start = self.resolution * int(segment.start / self.resolution + 0.5)
|
|
new_end = self.resolution * int(segment.end / self.resolution + 0.5)
|
|
new_segment = Segment(start=new_start, end=new_end)
|
|
new_annotation[new_segment, new_track] = label
|
|
|
|
support = current_file["annotated"].extent()
|
|
return new_annotation.support().crop(support)
|
|
|
|
|
|
class DeriveMetaLabels:
|
|
"""Filters out classes not defined in the `classes` list and
|
|
synthesizes additional classes based on unions or intersection of classes.
|
|
|
|
Parameters
|
|
----------
|
|
classes : List[str]
|
|
All the "base" classes that should be used in the protocol's annotation's
|
|
unions: Dict[str, List[str]], optional
|
|
Unions of classes. The keys are the name of the new union classes, and the values are the
|
|
list of classes that should used for these unions.
|
|
intersections: Dict[str, List[str]], optional
|
|
Intersections of classes. The keys are the name of the new intersections classes, and the values are the
|
|
list of classes that should used for these intersections.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
classes: list[str],
|
|
unions: dict[str, list[str]] | None = None,
|
|
intersections: dict[str, list[str]] | None = None,
|
|
):
|
|
self.classes: set[str] = set(classes)
|
|
self.unions = unions if unions is not None else {}
|
|
self.intersections = intersections if intersections is not None else {}
|
|
|
|
@property
|
|
def all_classes(self) -> list[str]:
|
|
"""A list of all the classes (base, union-based and intersection-based) that can be found
|
|
in output annotations from this preprocessor"""
|
|
return sorted(
|
|
list(self.classes)
|
|
+ list(self.unions.keys())
|
|
+ list(self.intersections.keys())
|
|
)
|
|
|
|
def __call__(self, current_file: ProtocolFile) -> Annotation:
|
|
annotation: Annotation = current_file["annotation"]
|
|
derived = annotation.subset(self.classes)
|
|
# Adding union labels
|
|
for union_label, subclasses in self.unions.items():
|
|
# creates a subset of the original annotation, based
|
|
mapping = {k: union_label for k in subclasses}
|
|
metalabel_annot = annotation.subset(subclasses).rename_labels(
|
|
mapping=mapping
|
|
)
|
|
derived.update(metalabel_annot.support())
|
|
|
|
# adding intersection labels
|
|
for intersect_label, subclasses in self.intersections.items():
|
|
# a bit trickier: for each intersection meta-class's subclass,
|
|
# we retrieve its timeline
|
|
subclasses_tl = [
|
|
annotation.label_timeline(subclass) for subclass in subclasses
|
|
]
|
|
# then we iteratively re-crop each of the timelines one onto
|
|
# the other
|
|
overlap_tl = reduce(lambda x, y: x.crop(y), subclasses_tl)
|
|
for seg in overlap_tl:
|
|
derived[seg] = intersect_label
|
|
|
|
return derived
|
|
|
|
|
|
class Waveform:
|
|
def __init__(self):
|
|
self._audio = Audio()
|
|
|
|
def __call__(self, file: ProtocolFile):
|
|
waveform, _ = self._audio(file)
|
|
return waveform
|
|
|
|
|
|
class SampleRate:
|
|
def __call__(self, file: ProtocolFile):
|
|
return get_audio_metadata(file).sample_rate
|