Speech Synthesis Based on Hidden Markov Models and Deep Learning

Abstract

Speech synthesis based on Hidden Markov Models (HMM) and other statistical parametric techniques have been a hot topic for some time. Using this techniques, speech synthesizers are able to produce intelligible and flexible voices. Despite progress, the quality of the voices produced using statistical parametric synthesis has not yet reached the level of the current predominant unit-selection approaches, that select and concatenate recordings of real speech. Researchers now strive to create models that more accurately mimic human voices. In this paper, we present our proposal to incorporate recent deep learning algorithms, specially the use of Long Short-term Memory (LSTM) to improve the quality of HMM-based speech synthesis. Thus far, the results indicate that HMM-voices can be improved using this approach in its spectral characteristics, but additional research should be conducted to improve other parameters of the voice signal, such as energy and fundamental frequency, to obtain more natural sounding voices.

Integrating Articulatory Information in Deep Learning-Based Text-to-Speech Synthesis

“…Deep learning models (e.g. deep neural network and variations) have been used for acoustic modeling in SPSS recently and outperformed traditional HMM-based approaches [11][12][13][14]. Despite the recent success of TTS, the conventional deep learning-based TTS still requires a large training data set from one speaker, which is a limit for some clinical application.…”