ASR System Routing via Affinity Status for Domain Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic speech recognition (ASR) systems are limited to specific subject matter domains, leading to recognition errors and communication breakdowns when attempting to connect with a wide range of domains using a single system.
Innovation Solution
An ASR system architecture that supports multiple recognition modules, an input-output controller manages affinity status to route utterances to specific modules based on prior recognition results or user requests, enhancing interaction with various domains through a client-server framework and decoupling conversation modules from speech recognition providers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single ASR system is used to connect to a wide range of subject matter domains, then the system's versatility is improved, but recognition errors increase and communication reliability deteriorates
Solution Approach 1:
The ASR system is divided into multiple specialized recognition modules, each trained for specific domains (e.g., medical, legal, technical). The controller segments incoming utterances and routes them to the most appropriate module based on domain identification, ensuring high recognition accuracy for domain-specific content while maintaining broad system versatility.
Solution Approach 2:
The controller acts as a universal coordinating component that manages multiple specialized recognition modules. It identifies the domain of incoming speech, selects the appropriate module, and routes the utterance accordingly. This multi-functional controller enables the system to handle diverse domains while maintaining high recognition accuracy through specialized module selection.
2Reliability
If all recognition modules process every speech recognition turn, then recognition accuracy is maintained, but processor load and system complexity increase
Solution Approach 1:
The system dynamically adjusts the recognition processing pathway based on the identified domain. The controller determines in real-time whether to route to a specialized module or use a general-purpose module, and whether affinity status should be established. This dynamic routing reduces system complexity by avoiding unnecessary processing while maintaining recognition accuracy through appropriate module selection.
Solution Approach 2:
The controller serves as an intermediary between the microphone input and multiple recognition modules. It mediates the routing decision by identifying domain keywords, selecting the most appropriate module, and managing affinity status. This intermediary approach simplifies the system architecture by centralizing the decision-making logic and preventing all modules from processing every utterance.
3Reliability
If multiple specialized recognition modules are used for different domains, then recognition accuracy for specific domains is improved, but system complexity and bandwidth requirements increase
Solution Approach 1:
Each recognition module is specialized with local quality optimized for its specific domain (e.g., medical terminology, legal jargon, technical specifications). The controller routes domain-specific utterances to modules with matching local expertise, ensuring high recognition accuracy for each domain while managing system complexity through targeted specialization rather than universal processing.
Solution Approach 2:
The system applies partial processing by routing only domain-specific utterances to specialized modules, while other utterances are handled by general-purpose modules or the same specialized module. Affinity status is established to avoid re-processing by multiple modules, reducing unnecessary bandwidth consumption and system complexity while maintaining high accuracy for domain-specific content.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system and method to receive a spoken utterance and convert the spoken utterance into a recognized speech results through multiple automatic speech recognition modules. Multiple conversation modules interpret the recognized speech results. The system and method assign an affinity status to one or more of the multiple automatic speech recognition modules. An affinity status restricts the conversion of a subsequent spoken utterance to a selected automatic speech recognition module or modules.