# Unsupervised word segmentation and lexicon discovery using acoustic word embeddings

arXiv: https://arxiv.org/abs/1603.02845 · HTML: https://mlsanity.com/paper/1603.02845/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 9 Mar 2016.

## Built on by (most cited first, 8 of 8)

- [Word Segmentation on Discovered Phone Units with Dynamic Programming and Self-Supervised Scoring](https://mlsanity.com/paper/2202.11929/) (2022)
- [A Masked Segmental Language Model for Unsupervised Natural Language Segmentation](https://mlsanity.com/paper/2104.07829/) (2021)
- [SpidR: Learning Fast and Stable Linguistic Units for Spoken Language Models Without Supervision](https://mlsanity.com/paper/2512.20308/) (2025)
- [Discovering Phonetic Inventories with Crosslingual Automatic Speech Recognition](https://mlsanity.com/paper/2201.11207/) (2022)
- [Towards Tokenized Human Dynamics Representation](https://mlsanity.com/paper/2111.11433/) (2021)
- [Unsupervised Multimodal Word Discovery based on Double Articulation Analysis with Co-occurrence cues](https://mlsanity.com/paper/2201.06786/) (2022)
- [Computational modeling of early language learning from acoustic speech and audiovisual input without linguistic priors](https://mlsanity.com/paper/2603.08359/) (2026)
- [DiscoPhon: Benchmarking the Unsupervised Discovery of Phoneme Inventories With Discrete Speech Units](https://mlsanity.com/paper/2603.18612/) (2026)

## Abstract (arXiv)

In settings where only unlabelled speech data is available, speech technology needs to be developed without transcriptions, pronunciation dictionaries, or language modelling text. A similar problem is faced when modelling infant language acquisition. In these cases, categorical linguistic structure needs to be discovered directly from speech audio. We present a novel unsupervised Bayesian model that segments unlabelled speech and clusters the segments into hypothesized word groupings. The result is a complete unsupervised tokenization of the input speech in terms of discovered word types. In our approach, a potential word segment (of arbitrary length) is embedded in a fixed-dimensional acoustic vector space. The model, implemented as a Gibbs sampler, then builds a whole-word acoustic model in this space while jointly performing segmentation. We report word error rates in a small-vocabulary connected digit recognition task by mapping the unsupervised decoded output to ground truth transcriptions. The model achieves around 20% error rate, outperforming a previous HMM-based system by about 10% absolute. Moreover, in contrast to the baseline, our model does not require a pre-specified vocabulary size.