Query-by-Example Spoken Term Detection (QbE-STD)

This repository implements a scalable Query-by-Example Spoken Term Detection (QbE-STD) system using:

• TF-IDF weighted k-mer similarity for efficient candidate retrieval.
• DBSCAN clustering for grouping similar audio token sequences.
• Smith-Waterman alignment for fine-grained sequence matching.
• wav2vec 2.0-based token extraction from audio.

The goal is to find audio segments that match a spoken query example.

---

✨ Features

• Extract token sequences from raw .wav audio using wav2vec2.
• Build TF-IDF weighted k-mer representations of audio token sequences.
• Cluster sequences automatically using DBSCAN (density-based clustering).
• Efficiently rank candidate matches based on cluster relevance.
• Perform fine matching using Smith-Waterman sequence alignment.