H-QuEST: Accelerating Query-by-Example Spoken Term Detection with Hierarchical Indexing

TL;DR

H-QuEST introduces a hierarchical indexing framework that significantly speeds up query-by-example spoken term detection using TF-IDF-based sparse representations and advanced audio features, maintaining high accuracy.

Contribution

The paper presents a novel hierarchical indexing approach combining TF-IDF representations and HNSW to accelerate spoken term detection without accuracy loss.

Findings

Substantial speed improvements in retrieval time.

Maintains high accuracy comparable to existing methods.

Effective scalability for large audio datasets.

Abstract

Query-by-example spoken term detection (QbE-STD) searches for matching words or phrases in an audio dataset using a sample spoken query. When annotated data is limited or unavailable, QbE-STD is often done using template matching methods like dynamic time warping (DTW), which are computationally expensive and do not scale well. To address this, we propose H-QuEST (Hierarchical Query-by-Example Spoken Term Detection), a novel framework that accelerates spoken term retrieval by utilizing Term Frequency and Inverse Document Frequency (TF-IDF)-based sparse representations obtained through advanced audio representation learning techniques and Hierarchical Navigable Small World (HNSW) indexing with further refinement. Experimental results show that H-QuEST delivers substantial improvements in retrieval speed without sacrificing accuracy compared to existing methods.