English

Generating Data with Text-to-Speech and Large-Language Models for Conversational Speech Recognition

Audio and Speech Processing 2024-08-20 v1 Computation and Language Sound

Abstract

Currently, a common approach in many speech processing tasks is to leverage large scale pre-trained models by fine-tuning them on in-domain data for a particular application. Yet obtaining even a small amount of such data can be problematic, especially for sensitive domains and conversational speech scenarios, due to both privacy issues and annotation costs. To address this, synthetic data generation using single speaker datasets has been employed. Yet, for multi-speaker cases, such an approach often requires extensive manual effort and is prone to domain mismatches. In this work, we propose a synthetic data generation pipeline for multi-speaker conversational ASR, leveraging a large language model (LLM) for content creation and a conversational multi-speaker text-to-speech (TTS) model for speech synthesis. We conduct evaluation by fine-tuning the Whisper ASR model for telephone and distant conversational speech settings, using both in-domain data and generated synthetic data. Our results show that the proposed method is able to significantly outperform classical multi-speaker generation approaches that use external, non-conversational speech datasets.

Keywords

Cite

@article{arxiv.2408.09215,
  title  = {Generating Data with Text-to-Speech and Large-Language Models for Conversational Speech Recognition},
  author = {Samuele Cornell and Jordan Darefsky and Zhiyao Duan and Shinji Watanabe},
  journal= {arXiv preprint arXiv:2408.09215},
  year   = {2024}
}

Comments

To appear at SynData4GenAI 2024 workshop

R2 v1 2026-06-28T18:15:32.428Z