Implements DelayedAggregation, a self-contained block that aggregates overlapping sliding-window outputs with configurable strategies (mean, hamming, first). This is a common pattern in streaming diarization and can be reused in any time-series streaming pipeline.

Provides SpeakerMap and SpeakerMapBuilder for aligning speaker labels across windows using distance, correlation, or MSE. This is a generic pattern for solving label permutation problems in clustering and can be extracted for other multi-object tracking or re-identification tasks.

Contains TemporalFeatureFormatter, which handles conversion between SlidingWindowFeature, numpy arrays, and torch tensors. This utility solves a common friction point in audio ML pipelines and can be reused in any project using pyannote-style features or sliding window data.

Defines custom RxPY operators for audio stream manipulation (e.g., buffer rearrangement, output accumulation). These operators demonstrate a clean pattern for building streaming data pipelines and can be reused in any RxPY-based audio or real-time system.

AudioLoader is a simple, self-contained utility for loading audio files with automatic resampling to a target sample rate. It wraps torchaudio and handles edge cases, making it reusable in any audio processing project.