TEXT-TO-SPEECH SYNTHESIS METHOD, DEVICE, COMPUTER APPARATUS, AND NON-VOLATILE COMPUTER READABLE STORAGE MEDIUM

摘要

A text-to-speech synthesis method, a device, and a computer apparatus. The text-to-speech synthesis method comprises: first acquiring a target text to be identified (101); performing discrete feature processing on each character in the target text to generate a corresponding feature vector for each character (102); inputting the feature vector into a pre-trained frequency spectrum conversion model, and acquiring a corresponding Mel-spectrum for each character in the target text, the Mel-spectrum being output by the frequency spectrum conversion model (103); and converting the Mel-spectrum into audio data to obtain audio data corresponding to the target text (104). Thus, speech synthesis is performed without generating phonemic notation of each character in a text so as to effectively reduce a workload during a speech synthesis process, provide an effective solution for pronunciation issues during the speech synthesis process, and achieve a wide application range in the field of artificial intelligence.