Artificial intelligence-based speech recognition method and deviceand storage medium
A speech recognition and artificial intelligence technology, applied in speech recognition, speech analysis, instruments, etc., can solve problems such as sensitivity to outliers, and achieve the effect of improving training efficiency and recognition accuracy
Pending Publication Date: 2021-12-31
PING AN TECH (SHENZHEN) CO LTD
View PDF0 Cites 4 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
However, since outliers usually have large gradients, this method has the problem of being very sensitive to outliers
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
[0059] It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.
[0060] In order to solve the problems existing in existing speech recognition schemes that only consider the probability of the most likely decoding path in the audio sample, but not all the decoding paths, and are sensitive to outliers, the present invention provides a method based on artificial Intelligent voice recognition method, the voice signal to be detected can include audio data of various lengths, after processing by the voice recognition module, the corresponding text information with the highest probability can be obtained, and then the recognition conversion from voice signal to text signal can be realized, and the recognition is accurate It has high accuracy, and the sample size required for model training is small, and the labeling cost is also low, which is suitable for various speech recognition scenar...
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 relates to artificial intelligence, and discloses an artificial intelligence-based speech recognition method, which comprises the steps of inputting acquired training data into a speech recognition module of a preset joint recognition model, and acquiring output data of the speech recognition module and first target task loss; inputting the output data into a loss prediction module of the joint recognition model to obtain second target task loss of the loss prediction module; based on the first target task loss and the second target task loss, acquiring the total task loss of the joint recognition model; performing iterative training on the joint recognition model based on the training data until the total task loss is converged in a preset range, and forming the joint recognition model; and recognizing a to-be-detected speech signal based on a speech recognition module in the joint recognition model, and obtaining a corresponding recognition result. According to the invention, the speech recognition precision and efficiency can be improved.
Description
technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to an artificial intelligence-based speech recognition method, device, electronic equipment and computer-readable storage medium. Background technique [0002] At present, end-to-end neural networks have gradually achieved remarkable results in automatic speech recognition (ASR) tasks. Among the existing models, CTC, encoder-decoder and CTC-attention hybrid architectures have attracted the attention of researchers at home and abroad. focus on. However, due to the complex network structure, these end-to-end models often face the problem of low computational efficiency, and these models also require a large amount of labeled speech data for training. As we all know, labeling speech data is an expensive and time-consuming task, and an audio often needs ten times the duration of the audio to be labeled. Therefore, optimizing training efficiency is necessary to ...
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 Applications(China)
IPC IPC(8): G10L15/06G10L15/32
CPCG10L15/063G10L15/32
Inventor 罗剑王健宗
Owner PING AN TECH (SHENZHEN) CO LTD