Sorry, you need to enable JavaScript to visit this website.

PROSODIC BOUNDARY PREDICTION MODEL FOR VIETNAMESE TEXT-TO-SPEECH

Primary tabs

Citation Author(s):
Nguyen Thi Thu Trang, Nguyen Hoang Ky, Albert Rilliard, Christophe D’Alessandro
Submitted by:
Trang Nguyen
Last updated:
22 October 2020 - 8:15am
Document Type:
Research Manuscript
Document Year:
2021
Event:
Presenters Name:
Nguyen Thi Thu Trang

Abstract 

Abstract: 

Prosodic boundary is a crucial prosodic cue of prosodic phrasing. This research aims to build a prosodic boundary prediction model for improving the naturalness of the Viet- namese speech synthesis. This model can be used directly to predict prosodic boundaries in synthesis phase of the statisti- cal parametric speech synthesis (e.g. Hidden Markov Model - HMM, Deep Neural Network - DNN). It can also be used to improve the quality of the training phase in the end-to- end speech synthesis (e.g. Tacotron). Beside a conventional feature of Part-Of-Speech (POS), the authors proposes two novel and efficient features to predict prosodic boundaries: syntactic blocks and syntactic links. Syntactic blocks are syn- tactic phrases whose sizes are bounded. The syntactic link of a word was a syntax tree-based relationship with the pre- vious word. These two important predictors are found based on a thorough analysis on VDTO Vietnamese corpus to find out a correlation between hierarchical syntactic information and pause appearance. The bounded size of syntactic blocks was discovered to be an optimal value of 10 syllables. The proposed features are experimented with the decision tree classification algorithm. The two novel predictors help the proposed model improve about 36.4% to the model with only POS features. The combination of all three predictors give the best F1-score results at 81.2% in 10-fold cross-validation and at 81.4% in test data.

up
0 users have voted:

Dataset Files

ICASSP_2021_Prosodic_Boundary_Prediction.pdf

(78)

ICASSP_2021_Prosodic_Boundary_Prediction.pdf

(56)

ICASSP_2021_Prosodic_Boundary_Prediction.pdf

(51)