Method for converting character into phonetic symbol

Abstract

The invention is a text-to-pronunciation converting method, comprising a morph-phoneme fragment find course and a three-stage text-to-pronunciation converting course, finding out possible morph-phoneme pairing sequence (called block) set by a ready-trained sound-marking dictionary, and in sequence making the three-stage text-to-pronunciation converting course of morph cutting, block marking and confirm program for the input letters so as to determine a pronouncing sequence. Reducing the phoneme chart capacity by marking with blocks of saving calculating resources largely increases the speed of searching possible block serials. Besides keeping accuracy above average level, it saves large amount of calculating time, and applies to speech related products on mobile information home applications.

Description

technical field [0001] The present invention relates to speech synthesis and speech recognition, in particular to a text-to-phonemisation method, which can be applied to a text-to-phonemisation model on mobile information appliances. Background technique [0002] Text-to-phonetic transcription is to convert the input text (text) into phonetic symbols (pronunciation) output. Long before the advent of the information appliance (IA) era, industry analysts around the world predicted that the voice human-machine interface would become the killer application of the information industry, and the technology of text-to-phonetic transcription (phonemisation) was often used in speech synthesis (speech synthesis) and speech recognition (speech recognition) related systems. [0003] Traditionally, the fastest way to obtain pronunciation is to look up a dictionary, but a dictionary cannot include all words and their pronunciations. Therefore, when the system encounters a new word that c...

AI Technical Summary

Problems solved by technology

[0012] Although the analog pronunciation has good performance in terms of execution efficiency, it is poor in accuracy

The joint N-gram model has good accuracy, but the graph established by the shape-meme-phoneme pairing (when n=4) is too large, resulting in the worst efficiency among the above methods

The multi-stage model will get higher pronunciation results. However, the method of reconfirming the phonetic elements that are prone to mislabeling not only increases the memory requirement, but also leads to the inability to effectively improve the overall efficiency.

Images