Inicio  /  Applied Sciences  /  Vol: 9 Par: 16 (2019)  /  Artículo
ARTÍCULO
TITULO

Exploring Efficient Neural Architectures for Linguistic?Acoustic Mapping in Text-To-Speech

Santiago Pascual    
Joan Serrà and Antonio Bonafonte    

Resumen

Conversion from text to speech relies on the accurate mapping from linguistic to acoustic symbol sequences, for which current practice employs recurrent statistical models such as recurrent neural networks. Despite the good performance of such models (in terms of low distortion in the generated speech), their recursive structure with intermediate affine transformations tends to make them slow to train and to sample from. In this work, we explore two different mechanisms that enhance the operational efficiency of recurrent neural networks, and study their performance?speed trade-off. The first mechanism is based on the quasi-recurrent neural network, where expensive affine transformations are removed from temporal connections and placed only on feed-forward computational directions. The second mechanism includes a module based on the transformer decoder network, designed without recurrent connections but emulating them with attention and positioning codes. Our results show that the proposed decoder networks are competitive in terms of distortion when compared to a recurrent baseline, whilst being significantly faster in terms of CPU and GPU inference time. The best performing model is the one based on the quasi-recurrent mechanism, reaching the same level of naturalness as the recurrent neural network based model with a speedup of 11.2 on CPU and 3.3 on GPU.

 Artículos similares

       
 
Jinxiong Gao, Xu Geng, Yonghui Zhang and Jingbo Wang    
Underwater autonomous path planning is a critical component of intelligent underwater vehicle system design, especially for maritime conservation and monitoring missions. Effective path planning for these robots necessitates considering various constrain... ver más
Revista: Applied Sciences

 
Yawei Ning, Minglei Ren, Shuai Guo, Guohua Liang, Bin He, Xiaoyang Liu and Rong Tang    
Multi-objective reservoir operation of reservoir flood control involves numerous factors and complex model solving, and exploring effective methods for solving the operation models has always been a hot topic in reservoir optimization operation research.... ver más
Revista: Water

 
Guo Li, Shuchun Huang, Wanqiu Lu, Junbo Liu, Shuiting Ding, Gong Zhang and Bo Zhen    
Probabilistic failure risk analysis of aeroengine life-limited parts is of great significance for flight safety. Current probabilistic failure risk analysis uses equal amplitude load calculations for conservative estimation, avoiding inclusion of the int... ver más
Revista: Aerospace

 
Kybeom Kwon, Seunghyun Min, Jongbum Kim and Kwangwon Lee    
The space mission analysis and design process defines a space system at the system level to accomplish space mission objectives. Although the traditional process is well established and comprehensive through several years of experience, we propose a nove... ver más
Revista: Aerospace

 
Youfei Hu, Haiyan Yang, Haolan Zhou and Qianwen Lv    
In the past decade, the numerical modelling of braided river morphodynamics has experienced a significant advance due to the increasing computer power and the development of numerical techniques. Numerical models are quite efficient in exploring scenario... ver más
Revista: Water