Adaptive Proper Name Entity Recognition in ASR Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic speech recognition (ASR) systems struggle with accurately recognizing proper names, especially uncommon ones, leading to incorrect results in natural language understanding applications.
Innovation Solution
The system employs a secondary recognizer capable of rapid adaptation to novel vocabulary words and proper names, using an adaptation object generation module to create specialized grammars for specific contexts, allowing for accurate recognition and understanding of proper names with low latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a standard open dictation ASR system is used, then the system can handle general speech recognition, but it fails to accurately recognize proper names and uncommon vocabulary
Solution Approach 1:
The system divides the speech recognition task into two segments: a first ASR system handles general speech recognition, while a second specialized ASR system handles proper names and uncommon vocabulary. This segmentation allows each system to be optimized for its specific function, resolving the contradiction between general versatility and specialized accuracy.
Solution Approach 2:
A language understanding module acts as an intermediary between the two ASR systems. It receives output from the first ASR system, identifies potential proper names, and directs those segments to the second ASR system for specialized processing. This intermediary enables the system to leverage both general speech recognition capabilities and specialized proper name recognition accuracy.
2Measurement precision
If extensive adaptation of open dictation ASR systems is performed to improve proper name recognition, then recognition accuracy improves, but computational cost and latency increase
Solution Approach 1:
The second ASR system is pre-configured with proper name dictionaries and specialized language models before runtime. This preliminary preparation eliminates the need for extensive adaptation during speech processing, reducing latency while maintaining high recognition accuracy for proper names.
Solution Approach 2:
The system extracts proper name recognition as a separate, specialized function from the general ASR system. By isolating this function into a dedicated second ASR system with specialized vocabulary, the system avoids the computational overhead of adapting a full general-purpose ASR system, thereby reducing latency and computational cost.
3Measurement precision
If extensive adaptation of ASR systems is performed, then proper name recognition improves, but computational resources and economic costs increase
Solution Approach 1:
Instead of uniformly enhancing the entire ASR system, the patent applies specialized proper name recognition resources locally where needed. The second ASR system contains only the necessary proper name dictionaries and language models specific to proper name recognition, rather than duplicating the full general-purpose ASR system. This localized approach reduces overall computational requirements while maintaining high accuracy for proper names.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Various embodiments contemplate systems and methods for performing automatic speech recognition (ASR) and natural language understanding (NLU) that enable high accuracy recognition and understanding of freely spoken utterances which may contain proper names and similar entities. The proper name entities may contain or be comprised wholly of words that are not present in the vocabularies of these systems as normally constituted. Recognition of the other words in the utterances in question— e.g., words that are not part of the proper name entities— may occur at regular, high recognition accuracy. Various embodiments provide as output not only accurately transcribed running text of the complete utterance, but also a symbolic representation of the meaning of the input, including appropriate symbolic representations of proper name entities, adequate to allow a computer system to respond appropriately to the spoken request without further analysis of the user's input.