- DEEP NEURAL NETWORK BASED NON-AUTOREGRESSIVE SPEECH SYNTHESIZER METHOD AND SYSTEM USING MULTIPLE DECODER

摘要

A deep neural network-based non-autoregressive speech synthesis method and system are presented. A deep neural network-based non-autoregressive speech synthesis system according to an embodiment configures a multi-decoder-based non-recursive deep neural network, and provides speech step by step through the multiple decoders from a template including temporal information of speech. a speech feature vector sequence synthesizing unit generating a feature vector sequence; and a speech reconstruction unit converting the speech feature vector sequence into speech data, wherein the speech feature vector sequence synthesizing unit generates the template input, and attaches an attention mechanism to the generated template input. ), by adding refined sentence data, a speech feature vector sequence can be generated.