r-G2P: Evaluating and Enhancing Robustness of Grapheme to Phoneme Conversion by Controlled noise introducing and Contextual information incorporation
This addresses robustness issues in G2P conversion for text-to-speech and speech recognition systems, representing an incremental improvement.
The paper tackled the sensitivity of neural grapheme-to-phoneme (G2P) models to spelling mistakes by proposing controlled noise introduction and contextual information incorporation, resulting in a robust model (r-G2P) that reduced word error rates by 2.73% on Dict-based benchmarks and 9.09% on real-world sources.
Grapheme-to-phoneme (G2P) conversion is the process of converting the written form of words to their pronunciations. It has an important role for text-to-speech (TTS) synthesis and automatic speech recognition (ASR) systems. In this paper, we aim to evaluate and enhance the robustness of G2P models. We show that neural G2P models are extremely sensitive to orthographical variations in graphemes like spelling mistakes. To solve this problem, we propose three controlled noise introducing methods to synthesize noisy training data. Moreover, we incorporate the contextual information with the baseline and propose a robust training strategy to stabilize the training process. The experimental results demonstrate that our proposed robust G2P model (r-G2P) outperforms the baseline significantly (-2.73\% WER on Dict-based benchmarks and -9.09\% WER on Real-world sources).