Methods for synthesizing vietnamese speech based on samples

Abstract

The invention relates to a method for synthesizing Vietnamese speech based on a template, including the following steps: Step 1: Normalizing the content of the template into written Vietnamese; Step 2: Checking the content of text segments as values ​​of individualized fields and text segments separated by individualized fields, provided that: the text consists of small text segments separated by individualized fields; and the individualized fields in the text all have accompanying values. Step 3: Calculate the acoustic characteristics in cases where the text does not have a corresponding audio file as described in Step 2, including pitch, duration, and intensity, then proceed to Step 4; Step 4: Find the optimal merging solution in cases where all sections of the text already have information on corresponding acoustic characteristics as described in Step 2 or as performed in Step 3; Step 5: Synthesize speech from the acoustic characteristics of the selected segments in the optimal splicing scheme saved in Step 4; Step 6: Splic and smooth, provided that all audio files corresponding to the segments in the text are available. The resulting artificial speech, synthesized according to the design, has a reading quality and naturalness equivalent to the synthesized speech from the entire text, with a synthesis speed of 3 to 5 seconds per 1000 characters.