Speech Recognition Modifying Discordant Named-Entities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition technologies face limitations in improving accuracy by modifying named-entities discordant with the parent domain in sentences, particularly in terms of computational efficiency and flexibility.
Innovation Solution
A method and device that modify named-entities discordant with the parent domain in a sentence by calculating probabilities based on word embeddings, using acoustic and language models to analyze pronunciation and context, and selecting candidate words with the highest adjusted probabilities to reduce errors and enhance recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical error correction process is used to produce candidate sentences, then speech recognition accuracy is improved, but processing complexity and computational resources increase
Solution Approach 1:
The patent segments the speech recognition process into distinct modules: acoustic model processing, language model processing, and named-entity recognition processing. Each module handles specific aspects independently, allowing for optimized computational resources while maintaining overall accuracy. The segmentation enables parallel processing of different candidate sentences without requiring full sequential analysis.
Solution Approach 2:
The patent applies local quality by focusing computational resources on specific critical areas rather than uniform processing. Named-entities discordant with parent domain are identified and processed with enhanced attention using domain-specific knowledge bases. This localized enhancement improves overall accuracy without proportionally increasing total computational complexity.
2Measurement precision
If word embeddings and probability calculations are used to modify named-entities, then speech recognition accuracy is improved, but computational time and processing speed may be reduced
Solution Approach 1:
The patent performs preliminary action by pre-calculating and storing word embedding representations and probability distributions for named-entities in advance. These pre-computed data structures are stored in memory for rapid retrieval during speech recognition processing. This eliminates the need for real-time computation of complex probability calculations, significantly improving processing speed while maintaining accuracy.
Solution Approach 2:
The patent applies partial action by selectively computing probability calculations only for named-entities that are identified as discordant with their parent domain. Rather than calculating probabilities for all words in the speech input, the system focuses computational resources only on problematic named-entities, reducing overall computational time while maintaining recognition accuracy for critical elements.
3Measurement precision
If domain-specific named-entity modification is implemented, then recognition accuracy in specific contexts is improved, but system adaptability to different domains decreases
Solution Approach 1:
The patent implements universality by designing a modular architecture where domain-specific knowledge bases can be dynamically loaded and switched. The core speech recognition engine remains domain-agnostic, while domain-specific named-entity databases and probability models can be activated based on the detected parent domain. This allows the system to adapt to multiple domains (weather, sports, entertainment, etc.) without sacrificing accuracy in any specific domain.
Solution Approach 2:
The patent applies dynamics by making the system's domain expertise flexible and changeable in real-time. The parent domain detection mechanism dynamically determines which domain-specific named-entity modification rules to apply during speech processing. This dynamic adaptation allows the system to switch between different domain configurations based on the input context, maintaining both specialization and versatility.
Data Source
AI summary
Disclosed are a speech recognition device and a speech recognition method which perform speech recognition by executing an artificial intelligence (AI) algorithms and/or a machine learning algorithm installed thereon, to communicate with other electronic devices and an external server in a 5G communication environment. The speech recognition method according to an embodiment of the present disclosure may include converting a series of spoken utterance signals to a text item, extracting a discordant named-entity that is discordant with a parent domain inferred form the text, calculating probabilities of candidate words associated with the discordant named-entity based on calculated distances between a term representing the parent domain and each candidate word associated with the discordant named-entity, and based on the calculated probabilities, modifying the discordant named-entity in the text to one of the candidate words associated with the discordant named-entity.