Unsupervised Word Segmentation using K Nearest Neighbors

TL;DR

This paper introduces an unsupervised kNN-based method for speech word segmentation that leverages pre-trained speech representations and outperforms existing single-stage approaches while competing with two-stage methods.

Contribution

It presents a novel unsupervised approach that directly uses pre-trained audio features and compares segments with their K nearest neighbors, eliminating the need for phoneme discovery.

Findings

Improved results over previous single-stage methods

Competitive performance with state-of-the-art two-stage methods

Operates directly on pre-trained speech representations

Abstract

In this paper, we propose an unsupervised kNN-based approach for word segmentation in speech utterances. Our method relies on self-supervised pre-trained speech representations, and compares each audio segment of a given utterance to its K nearest neighbors within the training set. Our main assumption is that a segment containing more than one word would occur less often than a segment containing a single word. Our method does not require phoneme discovery and is able to operate directly on pre-trained audio representations. This is in contrast to current methods that use a two-stage approach; first detecting the phonemes in the utterance and then detecting word-boundaries according to statistics calculated on phoneme patterns. Experiments on two datasets demonstrate improved results over previous single-stage methods and competitive results on state-of-the-art two-stage methods.