g2pW: A Conditional Weighted Softmax BERT for Polyphone Disambiguation in Mandarin
This addresses a key challenge in Mandarin text-to-speech systems, offering an incremental improvement by eliminating the need for separate POS tagging models.
The paper tackles polyphone disambiguation in Mandarin grapheme-to-phoneme conversion by proposing g2pW, a method that uses learnable softmax-weights conditioned on characters and POS tags, which outperforms existing methods on the CPP dataset.
Polyphone disambiguation is the most crucial task in Mandarin grapheme-to-phoneme (g2p) conversion. Previous studies have approached this problem using pre-trained language models, restricted output, and extra information from Part-Of-Speech (POS) tagging. Inspired by these strategies, we propose a novel approach, called g2pW, which adapts learnable softmax-weights to condition the outputs of BERT with the polyphonic character of interest and its POS tagging. Rather than using the hard mask as in previous works, our experiments show that learning a soft-weighting function for the candidate phonemes benefits performance. In addition, our proposed g2pW does not require extra pre-trained POS tagging models while using POS tags as auxiliary features since we train the POS tagging model simultaneously with the unified encoder. Experimental results show that our g2pW outperforms existing methods on the public CPP dataset. All codes, model weights, and a user-friendly package are publicly available.