An Efficient Speech Synthesizer – A Hybrid Monotonic architecture for text-to-speech using VAE & LPC-net with independent sentence length

Abstract

In this research, it is suggested that a hybrid architecture for text-to-speech, which is named it as Efficient Speech Synthesizer. ESS optimizes all the parameters through a consistent, end-to-end training process that enables fast and efficient speech synthesis of exceptional quality without depending on the length of the sentences. ESS is powered by combining feed-forward approaches like Variational Auto Encoders and Linear Predictive Coding techniques. These techniques give monotonic limitations to the sequence alignment with almost no increase in computation. The present research focuses on TTS model with low complexity, low latency which is suitable for both high-end and low-end computational devices like IoT and smart Phones.