Filter out too short segments from diarization in the separation pipeline

[clement-pages]

No description provided.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[Copilot]

Pull Request Overview

This PR adds filtering functionality to remove speech segments that are too short in the speaker diarization component of the speech separation pipeline.

• Adds a min_duration_on parameter to control the minimum duration of speech segments
• Implements morphological closing operation to filter out segments shorter than the threshold
• Updates import statements to include the necessary scipy.ndimage functions

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

[nitpick] The import reorganization places scipy.ndimage import before torch import, breaking alphabetical ordering. Consider maintaining consistent import ordering for better code organization.

Suggested change

	from scipy.ndimage import binary_dilation, binary_closing
	import torch
	import torch
	from scipy.ndimage import binary_dilation, binary_closing

[Copilot]

Creating a new numpy array with np.array([[True] * min_frames_on]).T for each call is inefficient. Consider creating the structure array once outside the conditional or reusing it across calls.

Suggested change

	discrete_diarization.data = binary_closing(
	discrete_diarization.data, structure=np.array([[True] * min_frames_on]).T
	)
	structure = np.array([[True] * min_frames_on]).T
	discrete_diarization.data = binary_closing(
	discrete_diarization.data, structure=structure
	)

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.