Speech Recognition Network Segmentation for Personalized Name Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional continuous speech recognition systems face challenges in accurately recognizing Chinese names due to limited training corpora, numerous homophones, and the lack of personalized name lists, resulting in poor recognition accuracy for user-specific contact names.
Innovation Solution
A speech recognition method and system that utilizes a basic static decoding network and an affiliated static decoding network to enhance recognition accuracy by extending the decoding network based on user-specific name language models, incorporating statistical probability models for common words and names, and extracting personalized name information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a basic static decoding network with a general name language model is used, then the system can handle common names, but the recognition accuracy for personalized user contact names deteriorates due to limited training corpus coverage
Solution Approach 1:
The decoding network is segmented into two distinct parts: a basic static decoding network for common names and a user-specific dynamic decoding network for personalized contact names. This segmentation allows each network to be optimized for its specific purpose, resolving the contradiction between general coverage and personalized accuracy.
Solution Approach 2:
The system transitions from a purely static decoding network to a dynamic architecture where a user-specific decoding network is constructed and integrated based on the current user's contact information. This dynamic adaptation enables the system to achieve high recognition accuracy for personalized names while maintaining the basic network's capability for common names.
2Device complexity
If a static decoding network with fixed language model is used, then the system structure remains simple, but the recognition accuracy deteriorates when encountering homophones and uncommon personalized names
Solution Approach 1:
The system performs preliminary action by pre-processing user contact information to extract name entities and their pronunciations before constructing the user-specific decoding network. This preliminary preparation enables the network to be optimized for the user's specific contact names, improving recognition accuracy for homophones and uncommon names without significantly increasing overall system complexity.
3Adaptability or versatility
If traditional continuous speech recognition is used, then the system can process general speech input, but the recognition accuracy deteriorates for name information due to polyphone characters and limited training data
Solution Approach 1:
The user-specific decoding network acts as an intermediary between the basic static decoding network and the final recognition output. When name recognition is detected, this intermediary network is activated to provide specialized processing for personalized contact names, resolving the contradiction between general speech processing capability and name recognition accuracy.
Data Source
Figure 1~2
Figure 3~4
Figure 5~7
AI summary
The present invention relates to a speech recognition method and system based on user personalized information. The method comprises the following steps: receiving a speech signal; decoding the speech signal according to a basic static decoding network to obtain a decoding path on each active node in the basic static decoding network, wherein the basic static decoding network is a decoding network associated with a basic name language model; if a decoding path enters a name node in the basic static decoding network, network extending is carried out on the name node according to an affiliated static decoding network of a user, wherein the affiliated static decoding network is a decoding network associated with a name language model of a particular user; and returning a recognition result after the decoding is completed. The recognition accuracy rate of user personalized information in continuous speech recognition may be raised by using the present invention.