Automatic classification of stop realisation with wav2vec2.0

2025-05-29

AI Summary

This study addresses the fine-grained phonological variation modeling challenge of automatic detection of stop bursts—a core phonetic phenomenon in speech science. We propose an end-to-end binary classification framework built upon wav2vec 2.0. Methodologically, we are the first to successfully adapt wav2vec 2.0 to millisecond-level stop realization discrimination, eliminating the need for phoneme alignment or hand-crafted feature engineering; instead, we directly leverage pretrained frame-level representations and attach a lightweight classifier head. Evaluated on cross-lingual (English and Japanese) and cross-quality (gold-standard vs. raw) corpora, our approach achieves high accuracy, with Pearson correlation coefficients exceeding 0.95 between automatically and manually annotated variant distributions. Our key contribution lies in overcoming the limitations of conventional tools in modeling variable phonological phenomena, enabling high-fidelity, scalable automatic annotation of large-scale speech corpora—thereby establishing a robust computational infrastructure for empirical research on phonological variation.

Technology Category

• Natural Language Processing: Speech
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Modern phonetic research regularly makes use of automatic tools for the annotation of speech data, however few tools exist for the annotation of many variable phonetic phenomena. At the same time, pre-trained self-supervised models, such as wav2vec2.0, have been shown to perform well at speech classification tasks and latently encode fine-grained phonetic information. We demonstrate that wav2vec2.0 models can be trained to automatically classify stop burst presence with high accuracy in both English and Japanese, robust across both finely-curated and unprepared speech corpora. Patterns of variability in stop realisation are replicated with the automatic annotations, and closely follow those of manual annotations. These results demonstrate the potential of pre-trained speech models as tools for the automatic annotation and processing of speech corpus data, enabling researchers to `scale-up' the scope of phonetic research with relative ease.

Problem

• Classifying stop burst presence using wav2vec2.0
• Enabling automatic annotation of variable phonetic phenomena
• Scaling phonetic research with pre-trained speech models

Innovation

• Uses wav2vec2.0 for stop classification
• Trains model for English and Japanese
• Replicates manual annotation accuracy