[PDF] Polyphone Disambiguation for Mandarin Chinese Using Conditional Neural Network with Multi-level Embedding Features | Semantic Scholar (2024)

[PDF] Semantic Reader

A novel system based on word-level features and window-based attention for polyphone disambiguation, which is a fundamental task for Grapheme-to-phoneme (G2P) conversion of Mandarin Chinese, is proposed.

This paper proposes a framework that can predict the pronunciations of Chinese characters, and the core model is trained in a distantly supervised way, and improves the predictive accuracy for unbalanced polyphonic characters.

A novel semi-supervised learning (SSL) framework for Mandarin Chinese polyphone disambiguation that can potentially leverage unlimited unlabeled text data and explores the effect of various proxy labeling strategies including entropy-thresholding and lexicon-based labeling.

This paper proposes a novel semi-supervised learning (SSL) framework for Mandarin Chinese polyphone disambiguation that can potentially leverage unlimited unlabeled text data and explores the effect of various proxy labeling strategies including entropy-thresholding and lexicon-based labeling.

This paper proposes a model that encodes both the input character sequence and dictionary matched words of the sentence, enabling the model to both avoid segment errors and leverage the well-designed pronunciation dictionary in the model.

A novel method to solve the problem of polyphone disambiguation when doing grapheme-to-phoneme (G2P) conversion in Mandarin Chinese text-to-speech systems and results show that it outperforms the existing methods on a public dataset called CPP.

A Chinese polyphone BERT model to predict the pronunciations of Chinese polyphonic characters is proposed, which can turn the polyphone disambiguation task into a pre-training task of the Chinese polyphones BERT.

The objective evaluation results showed that the proposed method improved the accuracy by 5.7 points in PD and 6.0 points in AP, and the perceptual listening test results confirmed that a TTS system employing the proposed model as a front-end achieved a mean opinion score close to that of synthesized speech with ground-truth pronunciation and accent in terms of naturalness.

A back-translation-style data augmentation method for mandarin Chinese polyphone disambiguation, utilizing a large amount of unlabeled text data, and a data balance strategy to improve the accuracy of some typical polyphonic characters in the training set with imbalanced distribution or data scarcity.

This work proposes 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 and shows that it outperforms existing methods on the public CPP dataset.

...

...

This paper proposes to use bidirectional long short-term memory (BLSTM) neural network to encode both the past and future observations on the character sequence as its inputs and predict the pronunciations of polyphonic characters.

This paper formulate the polyphone disambiguation problem into a classification problem and proposes a language independent classifier based on maximum entropy to address the issue, and introduces inequality smoothing to alleviate data sparseness and exploit language independent character features as linguistic knowledge.

A maximum entropy model to disambiguate polyphones, and various keyword selection approaches in different domains are proposed, and a hierarchical clustering algorithm for automatic generation of feature templates is designed, which minimizes the need for human supervision during ME model training.

The paper addresses the ambiguity issue of Chinese character polyphones and disambiguity approaches for such issues and proposed the unify approaches to improve the performance with respect to various threshold value.

This paper points out that correct G2P conversion for 41 key polyphones and 22 key polyphonic multi-syllabic words will constrain the overall error rate to below 0.068%, and proposes a semi-automatic approach to do this, which saves almost half of the workload.

This work proposes a G2P model based on a Long Short-Term Memory (LSTM) recurrent neural network (RNN) that has the flexibility of taking into consideration the full context of graphemes and transform the problem from a series of grapheme-to-phoneme conversions to a word- to-pronunciation conversion.

Direction Skip-gram (DSG), a simple but effective enhancement of the skip-gram model by explicitly distinguishing left and right context in word prediction, and which outperforms others on different datasets in semantic and syntactic evaluations.

A study on the relation between Chinese characters and their pronunciation, the solution to the disambiguation of polyphonic characters, a dictionary- based method, and a rules-based method are proposed.

The listening test results show that LSP and its dynamic counterpart, both in time and frequency, are preferred for the resultant higher synthesized speech quality.

...

...