Synthetic Data Domain Adaptation for ASR via LLM-based Text and Phonetic Respelling Augmentation
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915870691819520 |
|---|---|
| author | Yamashita, Natsuo Nagatsuka, Koichi Kokubo, Hiroaki Dohi, Kota Ho, Tuan Vu |
| author_facet | Yamashita, Natsuo Nagatsuka, Koichi Kokubo, Hiroaki Dohi, Kota Ho, Tuan Vu |
| contents | End-to-end automatic speech recognition often degrades on domain-specific data due to scarce in-domain resources. We propose a synthetic-data-based domain adaptation framework with two contributions: (1) a large language model (LLM)-based text augmentation pipeline with a filtering strategy that balances lexical diversity, perplexity, and domain-term coverage, and (2) phonetic respelling augmentation (PRA), a novel method that introduces pronunciation variability through LLM-generated orthographic pseudo-spellings. Unlike conventional acoustic-level methods such as SpecAugment, PRA provides phonetic diversity before speech synthesis, enabling synthetic speech to better approximate real-world variability. Experimental results across four domain-specific datasets demonstrate consistent reductions in word error rate, confirming that combining domain-specific lexical coverage with realistic pronunciation variation significantly improves ASR robustness. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_16920 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Synthetic Data Domain Adaptation for ASR via LLM-based Text and Phonetic Respelling Augmentation Yamashita, Natsuo Nagatsuka, Koichi Kokubo, Hiroaki Dohi, Kota Ho, Tuan Vu Audio and Speech Processing Sound End-to-end automatic speech recognition often degrades on domain-specific data due to scarce in-domain resources. We propose a synthetic-data-based domain adaptation framework with two contributions: (1) a large language model (LLM)-based text augmentation pipeline with a filtering strategy that balances lexical diversity, perplexity, and domain-term coverage, and (2) phonetic respelling augmentation (PRA), a novel method that introduces pronunciation variability through LLM-generated orthographic pseudo-spellings. Unlike conventional acoustic-level methods such as SpecAugment, PRA provides phonetic diversity before speech synthesis, enabling synthetic speech to better approximate real-world variability. Experimental results across four domain-specific datasets demonstrate consistent reductions in word error rate, confirming that combining domain-specific lexical coverage with realistic pronunciation variation significantly improves ASR robustness. |
| title | Synthetic Data Domain Adaptation for ASR via LLM-based Text and Phonetic Respelling Augmentation |
| topic | Audio and Speech Processing Sound |
| url | https://arxiv.org/abs/2603.16920 |