Speech Routing Architecture for Intent Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The selection of an appropriate application (skill) for processing user requests in speech processing systems is a non-trivial task due to the large number of potential applications and actions, making it challenging to determine the best application for fulfilling user intents efficiently.
Innovation Solution
The implementation of a speech processing routing architecture that uses machine learning models trained with CFIR tags, user feedback, and contextual data to dynamically rank and select the most suitable application for processing user requests, ensuring improved user experiences and accurate routing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large number of applications are available to process user requests, then the system's versatility and capability to fulfill user intents improve, but the difficulty of selecting the appropriate application increases
Solution Approach 1:
The patent introduces an intermediary component (routing system with machine learning models) that mediates between the user request and the large number of available applications. This intermediary ranks applications based on relevance to the user intent, effectively managing the complexity of selection while preserving access to diverse applications.
Solution Approach 2:
The system changes the parameter of application selection from manual or simple rule-based matching to machine learning-based ranking. By transforming the selection criterion into a ranked list based on learned patterns from CFIR tags and user feedback, the system efficiently navigates the large number of applications.
2Measurement precision
If traditional routing methods are used to select applications, then the system simplicity is maintained, but the routing accuracy and user satisfaction decrease
Solution Approach 1:
The system implements feedback loops where user interactions and outcomes are captured and used to retrain machine learning models. This feedback mechanism continuously improves routing accuracy by learning from actual user behavior and preferences, overcoming the limitations of static traditional routing methods.
Solution Approach 2:
The system performs preliminary ranking of applications before the actual request processing occurs. By pre-ranking applications based on historical data and CFIR tags, the system prepares the most relevant applications in advance, improving routing accuracy without adding significant complexity to the real-time processing.
3Measurement precision
If machine learning models are implemented to dynamically rank applications, then the routing accuracy and user satisfaction improve, but the computational resources and processing time increase
Solution Approach 1:
The system performs computationally intensive machine learning inference in advance by pre-ranking applications before the actual request processing. This preliminary ranking reduces the computational burden during real-time operations, as the heavy lifting of model evaluation is done beforehand when resources are more readily available.
Solution Approach 2:
The system applies different levels of processing to different applications based on their relevance. Highly relevant applications (top of the ranked list) receive more detailed evaluation, while less relevant applications are quickly filtered out. This local quality approach optimizes computational resources by focusing intensive processing only where necessary.
Data Source
AI summary
Devices and techniques are generally described for a speech processing routing architecture. First input data representing an input request may be received. First data may be sent to a first skill representing a first request for the first skill to evaluate an ability of the first skill to process the first input data. Second data may be sent to a second skill representing a second request for the second skill to evaluate an ability of the second skill to process the first input data. Third data may be received from the first skill indicating a first action performed by the first skill in response to receipt of the first input data. Fourth data may be received from the second skill indicating a second action performed by the second skill. The first skill may be selected for processing the first input data.