Prosodic hierarchy model training method, text-to-speech method and text-to-speech device

Abstract

The invention discloses a prosodic hierarchy model training method for text-to-speech and a text-to-speech method and a text-to-speech device by a prosodic hierarchy model. The training method includes: training massive non-tagged corpus data to obtain character vectors of individual characters; obtaining textual features and tagging corresponding to the trained data according to the character vectors and prosodic tagging data; based on a deep neural network and a bidirectional LSTM (long-short term memory) neural network, training the prosodic hierarchy model according to the textual features and tagging of the trained data. The training method has the advantages that when a character granularity based dictionary in the training method is compared with a traditional word granularity based dictionary, entry scale is decreased effectively, requirements of the model and resource files on computing resources and memory spaces are lowered, and usability of a prosodic prediction model in embedded intelligent devices is guaranteed while performance of the prosodic prediction model is improved.

Description

technical field [0001] The invention relates to the field of speech technology, in particular to a prosody-level model training method for speech synthesis, a method and a device for speech synthesis using the prosody-level model. Background technique [0002] Speech synthesis, also known as text-to-speech technology, is a technology that can convert text information into speech and read it aloud. In the speech synthesis system, since prosodic level prediction is the basis of the whole system, the key to how to improve the effect of speech synthesis is how to improve the accuracy of prosodic level prediction. [0003] In related technologies, there are mainly two methods for prosodic level prediction: [0004] First, the prosodic level prediction usually uses the CRF (Conditional Random Field, conditional random field) model for prediction, that is, the prosodic level prediction method based on CRF needs to expand the training features left and right in order to introduce c...

AI Technical Summary

Problems solved by technology

[0007] (1) Although the method of feature left and right expansion can introduce contextual relations to a certain extent, in order to reduce the scale of the model and the complexity of training, the size of the expansion is often limited, so it is impossible to build a long-distance contextual relationship between words;

[0008] (2) Adopting the method of level-by-level training may lead to wrong level-by-level transmission. Once an error occurs in the prosody prediction of the previous level, this error is easily passed down, causing subsequent prediction errors;

[0009] (3) Since the training and prediction of the prosody prediction model is based on word granularity, the performance of the prosody prediction model depends on the performance of the word segmentation system. Due to the limitation of computing resources and storage space, the performance of the word segmentation system in offline speech synthesis is limited. It is lower than the word segmentation system in the online speech synthesis system, thus affecting the final prosody prediction performance;

[0010] (4) For the offline synthesis system, due to the limited computing resources and storage space, there are strict requirements on the size of the model and resource files. The prediction model using word granularity needs to rely on hundreds of thousands of dictionary file entries. For storage space and take up a lot of computing resources

Images