SeriesIntroRecognizer ‐ Wiki

Overview

GPU-accelerated detection of recurring intros/outros across episodes. See installation, configuration, and a four-step pipeline from audio prep to final intervals.

Quick Start

• Install and verify GPU, then pip install series_intro_recognizer. See Installation.
• Tune key parameters like precision_secs, series_window, and segment bounds in Configuration.
• Run examples in Usage.

Pipeline

• Step 1 — Audio Data Preparation
  Decode, resample to a unified rate, mono-convert, normalize, apply optional (offset, duration), and drop too-short signals.
  Read more →

• Step 2 — Correlation and Offset Detection
  Pairwise async scan for coarse matches, then sync refinement for precise offsets; collect per-pair offset tuples.
  Read more →

• Step 3 — Best Offset Finder
  Consolidate many candidate intervals via NaN filtering, “close-values” fast path, then KMeans with silhouette; return a single Interval(start, end).
  Read more →

• Step 4 — Interval Construction and Improvement
  Assemble the longest correlated run around the best interval, enforce bounds, adjust edges, map to episodes, and deduplicate.
  Read more →

Core Docs

• Installation — GPU setup, CuPy variants, verification, common issues.
  Open →
• Configuration — User-facing parameters with effects and interactions.
  Open →
• Usage — File- and sample-based entry points, inputs/outputs, important warnings.
  Open →

Navigation

• Use the pipeline in order: Step 1 → Step 2 → Step 3 → Step 4.
• For errors or environment issues, start with Installation and Usage pages.