A Mandarin Prosodic Boundary Prediction Model Based on Multi-Task Learning

By Huashan PAN, Xiulin Li. 2019

20 citations recorded by the source corpus.

In this paper, we propose a mandarin prosodic boundary prediction model based on Multi-Task Learning (MTL) architecture. The prosody structure of mandarin is a three-level hierarchical structure, which contains three basic units--Prosodic Word (PW), Prosodic Phrase (PPH) and Intonational Phrase (IPH) [1]. Previous studies usually decompose mandarin prosodic boundary prediction task into three independent tasks on these three unit boundaries [1-4]. In recent years, with the development of deep learning, MTL has achieved state-of-the-art performance on many tasks in Natural Language Processing (NLP) field [5-7]. Inspired by this, this paper implements an MTL framework with Bidirectional Long-Short Term Memory and Conditional Random Field (BLSTM-CRF) as the basic model, and takes three independent tasks of mandarin prosodic boundary prediction as sub-modules for PW, PPH and IPH individually. Under the MTL architecture, the three independent tasks are unified for overall optimization. The experiment results show that our model is effective in solving the task of mandarin prosodic boundary prediction, in which the overall prediction performance is improved by 0.8%, and the model size is reduced by about 55%.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Xu Tan
2. Tao Qin
3. Tie-Yan Liu
4. Frank K. Soong
5. Xinyu Yang
6. Zhaoxi Mu
7. Yizhuo Dong
8. Sheng Zhao

Citations in the corpus

• A Survey on Neural Speech Synthesis (2021, methodology): “TTS Text Analysis Char→Linguistic TN [310, 223, 430], G2P [403, 320, 321] Prosody Prediction [312, 137, 277, 251] Unified Model [252, 437] DeepVoice 1/2 [8, 84]”
• Review of end-to-end speech synthesis technology based on deep learning (2021, methodology): “[154] modeled prosody information at all levels of the text in the way of multi-task learning, and proposed a Mandarin prosodic boundary prediction model based on BLSTM-CRF, which improved the prediction accuracy and simplified the model.”
• A Unified Sequence-to-Sequence Front-End Model for Mandarin Text-to-Speech Synthesis (2019, background): “Recently, Multi-Task Learning (MTL) architecture is also applied in prosody prediction [8].”
• Unified Mandarin TTS Front-end Based on Distilled BERT Model (2020, methodology): “, syntactic trees based rules for PSP [3], dictionary matching et al based polyphone disambiguity [4]), or statistical learning models (like CRF [5] and LSTM/Attention based NN models [6, 7] for PSP, [8, 9] for Mandarin G2P).”