Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

feat: add support for powerset in VAD and OSD pipelines #1467

Merged
merged 2 commits into from
Sep 22, 2023
Merged

feat: add support for powerset in VAD and OSD pipelines #1467

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ed62899
feat: add support for powerset in VAD and OSD pipelines
hbredin Sep 22, 2023
bba133d
Merge branch 'develop' into feat/pipeline-powerset-support
hbredin Sep 22, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 15 additions & 4 deletions pyannote/audio/pipelines/overlapped_speech_detection.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -134,9 +134,13 @@ def __init__(
)[:, :, -2, np.newaxis]
self._segmentation = Inference(model, **inference_kwargs)

#  hyper-parameters used for hysteresis thresholding
self.onset = Uniform(0.0, 1.0)
self.offset = Uniform(0.0, 1.0)
if model.specifications.powerset:
self.onset = self.offset = 0.5

else:
#  hyper-parameters used for hysteresis thresholding
self.onset = Uniform(0.0, 1.0)
self.offset = Uniform(0.0, 1.0)

# hyper-parameters used for post-processing i.e. removing short overlapped regions
# or filling short gaps between overlapped regions
Expand All @@ -152,14 +156,21 @@ def __init__(
self.recall = recall

def default_parameters(self):
# parameters optimized on DIHARD 3 development set
if self.segmentation == "pyannote/segmentation":
# parameters optimized on DIHARD 3 development set
return {
"onset": 0.430,
"offset": 0.320,
"min_duration_on": 0.091,
"min_duration_off": 0.144,
}

elif self.segmentation == "pyannote/segmentation-3.0.0":
return {
"min_duration_on": 0.0,
"min_duration_off": 0.0,
}

raise NotImplementedError()

def classes(self):
Expand Down
20 changes: 15 additions & 5 deletions pyannote/audio/pipelines/voice_activity_detection.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,29 +121,40 @@ def __init__(

# load model and send it to GPU (when available and not already on GPU)
model = get_model(segmentation, use_auth_token=use_auth_token)

inference_kwargs["pre_aggregation_hook"] = lambda scores: np.max(
scores, axis=-1, keepdims=True
)
self._segmentation = Inference(model, **inference_kwargs)

#  hyper-parameters used for hysteresis thresholding
self.onset = Uniform(0.0, 1.0)
self.offset = Uniform(0.0, 1.0)
if model.specifications.powerset:
self.onset = self.offset = 0.5
else:
#  hyper-parameters used for hysteresis thresholding
self.onset = Uniform(0.0, 1.0)
self.offset = Uniform(0.0, 1.0)

# hyper-parameters used for post-processing i.e. removing short speech regions
# or filling short gaps between speech regions
self.min_duration_on = Uniform(0.0, 1.0)
self.min_duration_off = Uniform(0.0, 1.0)

def default_parameters(self):
# parameters optimized on DIHARD 3 development set
if self.segmentation == "pyannote/segmentation":
# parameters optimized for DIHARD 3 development set
return {
"onset": 0.767,
"offset": 0.377,
"min_duration_on": 0.136,
"min_duration_off": 0.067,
}

elif self.segmentation == "pyannote/segmentation-3.0.0":
return {
"min_duration_on": 0.0,
"min_duration_off": 0.0,
}

raise NotImplementedError()

def classes(self):
Expand Down Expand Up @@ -289,7 +300,6 @@ def __init__(
self.learning_rate = LogUniform(1e-6, 1)

def apply(self, file: AudioFile) -> Annotation:

# create a copy of file
file = dict(file)

Expand Down
Loading