Adversarial Training For Low-Resource Disfluency Correction

Vineet Bhat, Preethi Jyothi, Pushpak Bhattacharyya
6/10/2023
cs.CL

No official code found. Best implementation: vineet2104/AdversarialTrainingForDisfluencyCorrection (3 stars, MIT, last commit 2023-07-14).

Abstract

Disfluencies commonly occur in conversational speech. Speech with disfluencies can result in noisy Automatic Speech Recognition (ASR) transcripts, which affects downstream tasks like machine translation. In this paper, we propose an adversarially-trained sequence-tagging model for Disfluency Correction (DC) that utilizes a small amount of labeled real disfluent data in conjunction with a large amount of unlabeled data. We show the benefit of our proposed technique, which crucially depends on synthetically generated disfluent data, by evaluating it for DC in three Indian languages- Bengali, Hindi, and Marathi (all from the Indo-Aryan family). Our technique also performs well in removing stuttering disfluencies in ASR transcripts introduced by speech impairments. We achieve an average 6.15 points improvement in F1-score over competitive baselines across all three languages mentioned. To the best of our knowledge, we are the first to utilize adversarial training for DC and use it to correct stuttering disfluencies in English, establishing a new benchmark for this task.

View on arXivView PDF

Code Implementations(1)

vineet2104/AdversarialTrainingForDisfluencyCorrectionCommunity100%
MIT

This repository contains the code for training a Seq-GAN-BERT adversarial training model for disfluency correction using labeled and unlabeled sentences.

31Python, Jupyter NotebookMay 23, 20233 years agoMIT

Similar Papers

Mind the Pause: Disfluency-Aware Objective Tuning for Multilingual Speech Correction with LLMs

87% match1 repo2026

Automatic Disfluency Detection from Untranscribed Speech

84% match1 repo362023

DISCO: A Large Scale Human Annotated Corpus for Disfluency Correction in Indo-European Languages

81% match1 repo42023

Adapting the NICT-JLE Corpus for Disfluency Detection Models

81% match2023

Neural POS Tagging for the Low-Resource Saraiki

81% match2026

Inclusive ASR for Disfluent Speech: Cascaded Large-Scale Self-Supervised Learning with Targeted Fine-Tuning and Data Augmentation

81% match2024

Cite this paper

@article{bhat2023adversarial,
  title  = {Adversarial Training For Low-Resource Disfluency Correction},
  author = {Vineet Bhat and Preethi Jyothi and Pushpak Bhattacharyya},
  year   = {2023},
  eprint = {2306.06384},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2306.06384v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2306.06384.svg)](https://codeofpaper.com/paper/2306.06384)

Discussion