A method, system, device and storage medium for converting text into speech
A technology for converting text to speech, applied in speech analysis, speech synthesis, instruments, etc., can solve the problems of high cost, restricting the development of text-to-speech technology, scarcity, etc., and achieve the effect of low cost
Active Publication Date: 2022-06-21
RES INST OF TSINGHUA PEARL RIVER DELTA +1
View PDF11 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
The perfect training sample means that the speech in the training sample comes from a plurality of different speakers, and each speaker speaks these speeches in different languages; and if the speech belonging to different languages comes from different speakers, the Due to the strong correlation between language and speaker identity, the quality of training samples is reduced and the desired training effect cannot be achieved.
In reality, even if there are not many people who are proficient in two languages, and those who are proficient in three or more languages are scarce, the cost of obtaining perfect training samples is extremely high, which restricts the development of existing text-to-speech technology , which makes it difficult for the existing technology to realize the function of converting text mixed in multiple different languages into speech at a low cost and controlling the speech style at will
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0049] In this embodiment, the principle of the method for converting text into speech is as follows figure 1 shown. The basic process is as follows: perform preprocessing such as amplitude normalization, trimming silence and spectral conversion of the speech to be migrated to obtain the linear spectrum or mel spectrum of the speech to be migrated, and then input the linear spectrum or mel spectrum into the speech style encoder. On the other hand, after preprocessing the test text such as sentence segmentation and word segmentation, it is input into the attention-based auto-encoding model, and the output from the auto-encoding model is obtained. Pronunciation coding: The style coding and pronunciation coding are spliced and input into the speech decoder, and the spectrum output by the speech decoder is obtained after processing, and then the spectrum is converted into the obtained speech.
[0050] refer to figure 1 , the voice style encoder is composed of a multi-layer two...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses a method, system, device and storage medium for converting text into speech. The method includes obtaining test text and speech to be transferred, obtaining the pronunciation code of the test text, and using a trained speech style coder , obtaining the style code of the speech to be transferred, concatenating the pronunciation code and the style code, and decoding the concatenated result by using a speech decoder to obtain the speech. By using a speech style encoder, the style of the speech to be transferred can be extracted, and then integrated into the pronunciation of the test text, so that the final speech has the same content as the test text and the same style as the speech to be transferred, achieving from Text-to-speech conversion with a certain style. Moreover, the present invention can obtain a training set at a relatively low cost for training the speech style encoder, thereby reducing the cost of use. The invention is widely used in the technical field of voice data.
Description
technical field [0001] The present invention relates to the technical field of speech data, in particular to a method, system, device and storage medium for converting text into speech. Background technique [0002] In the fields of computer and news dissemination, there is a widespread need to convert texts that are automatically generated by computers or externally input to computers into fluent and easy-to-understand voice information. The existing text-to-speech technology (TTS technology) is mainly implemented based on an end-to-end neural network model, which can control the style of the output voice from the same text. However, if this advantage is to be extended to multiple The support of different languages relies on the use of well-established training samples to train the neural network model. The perfect training sample means that the voices in the training samples come from multiple different speakers, and each speaker speaks these voices in different languag...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G10L13/02G10L13/033G10L25/24G10L25/03G10L25/30G10L25/60
CPCG10L13/02G10L13/033G10L25/24G10L25/03G10L25/30G10L25/60
Inventor 叶俊杰李权王伦基黄桂芳任勇韩蓝青
Owner RES INST OF TSINGHUA PEARL RIVER DELTA