Large-scale speech recognition method and system capable of quickly updating language model
A language model and speech recognition technology, applied in speech recognition, speech analysis, instruments, etc., can solve the problems of high cost, training language model time, large computing space, large storage space, inability to accurately recognize the latest vocabulary, etc., to achieve guaranteed recognition rate, the effect of improving the recognition rate
Active Publication Date: 2022-03-25
AISPEECH CO LTD
View PDF0 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
A large amount of training text leads to a very large amount of time, computing space, and storage space required for training a language model, so the update cycle of this large-scale language model is usually several weeks or even longer
However, users tend to speak some of the latest and most popular words recently, and the language model with an update cycle of several weeks or even longer cannot accurately identify the latest words. If the language model is updated frequently, the cost will be too high
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
[0039] In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
[0040] In the following, the embodiments of the present application will be introduced first, and then experimental data will be used to verify the differences between the solutions of the present application and the prior art, and what beneficial effects can be achieved.
[0041] Please refer to image 3 , which shows a flow chart...
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 large-scale speech recognition method capable of rapidly updating a language model and a system method comprising: acquiring a plurality of statement structure sentences. Obtain entries corresponding to multiple categories. Extended speech to text. Select a plurality of category entries to be updated from the plurality of category entries in the speaking structure sentence. Train entry language models for various new entries. Update entries. A decoding network is compiled according to the language model of the utterance and the language model of the entry in combination with the acoustic model. Get the recognition result. The present invention combines the traditional language model and the class-based language model to provide a large-scale speech recognition system that can be quickly updated, and can train a speech recognition system that includes only a few MB of computing and storage space within a few minutes. The language model of the latest vocabulary improves the recognition rate of new vocabulary while ensuring the recognition rate of the original large vocabulary.
Description
technical field [0001] The invention belongs to the technical field of speech recognition, and in particular relates to a large-scale speech recognition method and system capable of rapidly updating a language model. Background technique [0002] With the wide application of speech recognition technology, more and more smart devices have the function of speech recognition, and more and more users use the function of speech recognition to realize interaction with the device. These smart devices are feature-rich and support a variety of conversational styles and domains, requiring a continuous speech recognition system that covers a very large vocabulary. [0003] In order to meet such speech recognition requirements, the specific implementation methods of traditional speech recognition systems are as follows: figure 1 , first design the user's statement, and obtain the user's entry, then expand the user's statement and the user's entry, and then use the expanded text to trai...
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): G10L15/00G10L15/06G10L15/08G10L15/26
CPCG10L15/063G10L15/005G10L15/08G10L15/26G10L2015/0635G10L2015/088
Inventor 陆沁
Owner AISPEECH CO LTD