Pronunciation bias error detection method and device and storage medium
A technology of misdetection and speech, which is applied in speech analysis, speech recognition, instruments, etc., can solve problems such as difficulty in building a pronunciation error detection system, and achieve the effect of improving performance
Active Publication Date: 2021-08-31
BEIJING LANGUAGE AND CULTURE UNIVERSITY
View PDF6 Cites 2 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0007] However, the error detection method mentioned in the above reference, in the absence of pronunciation training data, although the target language L2 corpus is obtained through a small-scale speech corpus, such as about 150 hours of pre-training, the error detection performance is improved. , but for pronunciation features such as Chinese learning foreign languages, with a large span of speech levels and significant acoustic differences, it is still difficult to build a robust and good pronunciation error detection system
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
[0037] Below, the present invention will be described in detail with reference to the accompanying drawings.
[0038]In the transfer learning theory of deep learning, for a certain signal, such as image text speech, it is hoped that through pre-training, the model can extract a general representation that reflects the internal structure of the signal. In this way, different tasks in the same domain can benefit from this common representation. Specific to specific tasks, it is feasible to directly use this general representation as a feature, or add task-specific modules to perform overall fine-tuning on the pre-trained model. The pre-training model refers to learning a general representation from a large-scale corpus and using it for downstream tasks.
[0039] The pre-training model wav2vec2.0 is an open source pre-training model that belongs to the wav2vec series. The pre-trained model wav2vec2.0 achieves SOTA (stata-of-the-art, frontier or highest level) performance on mul...
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 provides a pronunciation bias error detection method and device and a storage medium. The method comprises the following steps: constructing a voice pre-training model, and pre-training the voice pre-training model based on an unlabeled voice corpus; adding a randomly initialized full connection layer on the voice pre-training model to obtain a fine tuning pre-training model, and training the fine tuning pre-training model by using annotated pronunciation bias error data to obtain a pronunciation bias error detection model; and detecting the voice of a learner by using the pronunciation bias error detection model to obtain pronunciation error information. According to the pronunciation bias error detection method and device and the storage medium, by constructing the voice pre-training model, finely tuning the pre-training model, the voice of the learner is detected by using the pronunciation bias error detection model so as to obtain the pronunciation bias error information, so that the performance of a pronunciation error detection system can be effectively improved in the absence of pronunciation training data.
Description
technical field [0001] The invention relates to the technical field of speech recognition, in particular to a method, device and storage medium for detecting pronunciation errors. Background technique [0002] With the development of speech technology and the promotion of online learning, Computer-Aided Pronunciation Training (CAPT) has been more and more used in language teaching. Among them, automatic pronunciation error detection is an important part of computer-aided pronunciation teaching. It is mainly used to detect learners' pronunciation errors, so as to help learners find and correct their pronunciation problems in the process of second language learning. [0003] The main principle of the pronunciation error detection technology is to obtain a pronunciation error detection system including all phoneme sets in the target language L2 through a large amount of target language L2 (Second / Targetlanguage, L2) speech corpus training. During detection, the corresponding p...
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 IPC IPC(8): G10L15/06G10L15/02G10L15/16G10L15/187G10L25/30G10L25/51
CPCG10L15/063G10L15/02G10L15/187G10L15/16G10L25/30G10L25/51G10L2015/025
Inventor 张劲松彭霖铠付凯奇解焱陆柯登峰
Owner BEIJING LANGUAGE AND CULTURE UNIVERSITY