wip: use hooks in clustering step

This commit is contained in:
Hervé BREDIN committed 2024-11-08 10:25:54 +01:00
1 parent 3f4a22f773
commit c7581cfa18
2 files changed
+13 -1

No files matched your search

+9 -1
View File
@@ -24,7 +24,7 @@
import random
from enum import Enum
from typing import Optional, Tuple
from typing import Callable, Optional, Tuple
import numpy as np
from einops import rearrange
@@ -209,6 +209,7 @@ class BaseClustering(Pipeline):
num_clusters: Optional[int] = None,
min_clusters: Optional[int] = None,
max_clusters: Optional[int] = None,
hook: Optional[Callable] = None,
**kwargs,
) -> np.ndarray:
"""Apply clustering
@@ -267,6 +268,7 @@ class BaseClustering(Pipeline):
min_clusters=min_clusters,
max_clusters=max_clusters,
num_clusters=num_clusters,
hook=hook,
)
hard_clusters, soft_clusters, centroids = self.assign_embeddings(
@@ -326,6 +328,7 @@ class AgglomerativeClustering(BaseClustering):
min_clusters: Optional[int] = None,
max_clusters: Optional[int] = None,
num_clusters: Optional[int] = None,
hook: Optional[Callable] = None,
):
"""
@@ -374,9 +377,13 @@ class AgglomerativeClustering(BaseClustering):
embeddings, method=self.method, metric=self.metric
)
hook("clustering.dendrogram", dendrogram)
# apply the predefined threshold
clusters = fcluster(dendrogram, self.threshold, criterion="distance") - 1
hook("clustering.fcluster", clusters)
# split clusters into two categories based on their number of items:
# large clusters vs. small clusters
cluster_unique, cluster_counts = np.unique(
@@ -470,6 +477,7 @@ class AgglomerativeClustering(BaseClustering):
# re-number clusters from 0 to num_large_clusters
_, clusters = np.unique(clusters, return_inverse=True)
return clusters
@@ -553,6 +553,7 @@ class SpeakerDiarization(SpeakerDiarizationMixin, Pipeline):
max_clusters=max_speakers,
file=file, # <== for oracle clustering
frames=self._segmentation.model.receptive_field, # <== for oracle clustering
hook=hook,
)
# hard_clusters: (num_chunks, num_speakers)
# centroids: (num_speakers, dimension)
@@ -590,6 +591,9 @@ class SpeakerDiarization(SpeakerDiarizationMixin, Pipeline):
# shape: (num_chunks, num_speakers)
hard_clusters[inactive_speakers] = -2
hook("hard_clusters", hard_clusters)
discrete_diarization = self.reconstruct(
segmentations,
hard_clusters,