Improving Mongolian Phrase Break Prediction by Using Syllable and Morphological Embeddings with BiLSTM Model

Abstract

In the speech synthesis systems, the phrase break (PB) prediction is the first and most important step. Recently, the state-of-the-art PB prediction systems mainly rely on word embeddings. However this method is not fully applicable to Mongolian language, because its word embeddings are inadequate trained, owing to the lack of resources. In this paper, we introduce a bidirectional Long Short Term Memory (BiLSTM) model which combined word embeddings with syllable and morphological embedding representations to provide richer and multi-view information which leverages the agglutinative property. Experimental results show the proposed method outperforms compared systems which only used the word embeddings. In addition, further analysis shows that it is quite robust to the Out-of-Vocabulary (OOV) problem owe to the refined word embedding. The proposed method achieves the state-of-the-art performance in the Mongolian PB prediction.

Exploiting Morphological and Phonological Features to Improve Prosodic Phrasing for Mongolian Speech Synthesis

“…Second, the nature of the specific model. Unlike traditional RNN-based sequence labeling models [62,79,80] that capture sequential context, self-attention sublayer in Figure 1 connects two arbitrary words directly regardless of their distance [53,58]. Furthermore, the recurrent sublayer in Figure 1 well captures the long-range sequential dependency, therefore, the self-attention layer in Figure 1 doesn't rely on an output layer to model the phrase break labeling sequence for decision making.…”

“…We adopt a self-attention neural classifier, which handles long range dependency of words better than RNN [52]. This work is an extension to our previous work [62] with several novel contributions,…”

Modeling Prosodic Phrasing With Multi-Task Learning in Tacotron-Based TTS

“…For Chinese, we use the Tencent AI Lab embedding database for Chinese Words and Phrases [41]. For Mongolian, the pre-trained 200-dimension word embedding reported in [42] is used.…”

Graphspeech: Syntax-Aware Graph Attention Network for Neural Speech Synthesis

“…In these techniques, the key idea is to integrate the conventional TTS pipeline into a unified encoder-decoder network and to learn the mapping directly from the <text, wav> pair. Tacotron is a successful encoder-decoder implementation based on recurrent neural networks (RNN), such as LSTM [11,12] and GRU [13]. However, the recurrent nature inherently limits the possibility of parallel computing in both training and inference.…”