User-guided ASR Arbitration via Multi-module Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic speech recognition (ASR) systems are limited by their inability to connect with a wide range of subject matter domains via a single system, leading to recognition errors and communication breakdowns.
Innovation Solution
A system that utilizes multiple automatic speech recognition modules, conversation modules, and an input-output ASR controller to receive spoken utterances, convert them into recognized speech results, collect ratings, and generate arbitration recommendations for user-guided selection, allowing for affinity status assignment and efficient processing of subsequent speech recognition turns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single ASR system is used, then the system complexity is low, but the ability to connect to wide range of subject matter domains is limited leading to recognition errors
Solution Approach 1:
The system is divided into multiple specialized ASR modules, each trained for specific subject matter domains. Instead of one general ASR system, the patent segments the recognition function across multiple domain-specific modules that can be selectively activated based on the conversation context.
Solution Approach 2:
The ASR controller provides a universal interface that can work with multiple different ASR modules across various domains. The controller manages multiple specialized modules through a single unified control mechanism, allowing the system to handle diverse domains while maintaining manageable complexity.
2Adaptability or versatility
If multiple ASR modules are used, then the coverage of subject matter domains improves, but the processor load and bandwidth increase
Solution Approach 1:
The system dynamically selects which ASR modules to activate based on the current conversation context and domain requirements. Rather than running all modules continuously, the controller adjusts the active module set in real-time, reducing processor load while maintaining comprehensive domain coverage when needed.
Solution Approach 2:
The ASR controller automatically manages the selection and coordination of multiple ASR modules without requiring external intervention. The system self-regulates which modules are active based on contextual cues, reducing the need for continuous high-level processing while maintaining domain adaptability.
3Measurement precision
If multiple ASR modules process speech simultaneously, then the recognition accuracy across domains improves, but the system response time increases
Solution Approach 1:
ASR modules are pre-configured and trained for specific domains before runtime. The controller pre-establishes which modules are relevant for given contexts, so when speech is received, the system can immediately activate the appropriate pre-prepared modules rather than initializing or selecting them in real-time, reducing response time while maintaining accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An automatic speech recognition (ASR) system includes at least one processor and a memory storing instructions. The instructions, when executed by the at least one processor cause the system to receive a spoken utterance and convert the spoken utterance into a plurality of recognized speech results through a plurality of automatic speech recognition modules; collect ratings for the recognized speech results by processing the recognized speech results through a plurality of conversation modules; determine that at least two of the collected ratings are equivalent and generate an arbitration recommendation indicating recognized speech results with equivalent collected ratings; and process the arbitration recommendation through one or more of the plurality of conversation modules for selection by one of the one or more of the plurality of conversation modules, wherein the one of the one or more of the plurality of conversation modules is adapted to allow user selection amongst the recognized speech results indicated by the arbitration recommendation. Related methods and computer-readable media are also disclosed.