Answers are grounded in this paper and cite their sources.
Attention mechanism is one of the most successful in deep learning based Natural Language (NLP). The transformer network architecture is based on attention mechanisms, and it sequence-to-sequence models in neural machine without recurrent and convolutional layers. -to-phoneme (G2P) conversion is a task of letters (grapheme sequence) to their (phoneme sequence). It plays a significant in text-to-speech (TTS) and automatic speech (ASR) systems. In this paper, we investigate the application of transformer architecture to G2P conversion and compare its performance with recurrent and neural network based approaches. Phoneme word error rates are evaluated on the CMUDict dataset US English and the NetTalk dataset. The results show that transformer based G2P outperforms the convolutional-based approach in terms of word error rate and our results significantly exceeded previous recurrent approaches (without attention) regarding word and phoneme error rates on both datasets. Furthermore, the size of the proposed model is much smaller than the size of the previous approaches.
No takes yet. Share an insight, caveat, or question.
Yolchuyeva et al. (2019) studied this question.