Context-Aware Intent Recognition Using Parallel ASR Nodes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated speech recognition (ASR) and intent recognition systems face limitations in performance and scalability, requiring extensive training data and effort to refine, and struggle with accurately transcribing and interpreting user intents across diverse contexts.
Innovation Solution
A context-aware speech processing architecture employing multiple nodes, each with intentionally limited speech recognition capabilities focused on specific contexts, processes voice inputs in parallel to extract user intent, using a confidence-based selection mechanism to determine the most relevant output.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single ASR system with extensive vocabulary and language model is used to handle diverse contexts, then the system can recognize a wider range of intents, but the complexity and computational burden increase significantly
Solution Approach 1:
The patent divides the intent recognition system into multiple specialized ASR engines, each configured with a specific lexicon for a particular context (e.g., restaurant ordering, flight booking, hotel reservation). This segmentation allows each engine to focus on a narrow domain, reducing individual complexity while collectively covering diverse intents through parallel processing of speech inputs across multiple engines.
2Measurement precision
If the lexicon and language model are extensively expanded to cover all possible contexts, then transcription accuracy improves across diverse inputs, but the training data requirements and refinement effort increase substantially
Solution Approach 1:
Instead of creating one universally accurate ASR system, the patent creates multiple ASR engines where each engine has a lexicon optimized for its specific context. For example, the restaurant ordering engine has a lexicon with food items, menu prices, and ordering terminology, while the flight booking engine has aviation-specific terminology. This local optimization achieves high transcription accuracy within each context without requiring extensive training data across all possible domains.
3Measurement precision
If multiple specialized ASR engines with context-specific lexicons are used, then transcription accuracy and intent recognition precision improve for specific contexts, but the device complexity and number of processing components increase
Solution Approach 1:
The system segments intent recognition into multiple specialized ASR engines, each handling a specific context with its own lexicon. This segmentation improves precision for each context while managing complexity through modular design where each engine is independently configured and can be added or removed based on required contexts.
Solution Approach 2:
Despite having context-specific lexicons, all ASR engines share a common underlying architecture and processing pipeline. The framework for receiving speech inputs, processing through the lexicon, and generating intent outputs is universal across all engines, reducing the effective complexity increase from having multiple specialized components.
4Adaptability or versatility
If conventional ASR systems attempt to transcribe all speech inputs accurately across diverse contexts, then overall system capability improves, but the processing time and computational resources required increase
Solution Approach 1:
By segmenting the ASR system into multiple specialized engines with context-specific lexicons, the system can quickly match speech inputs against relevant vocabulary without searching through an exhaustive universal lexicon. This segmentation improves processing speed for each context while maintaining versatility through the collective capability of multiple engines.
Data Source
AI summary
A computing system for determining a user intent from a speech input to effect a user intended action is provided. The computer system comprises a set of processing nodes and a controller module. Each processing node is capable of understanding only a subset of words directly relevant to a particular context. The processing nodes of the set are arranged to receive a same speech input, and each processing node attempts to interpret the input, based on its subset of words, to extract therefrom an output indicative of user intent. Each node is unable to interpret any portion of the input containing a word outside of its subset. The controller module receives the outputs from the set of processing nodes and determine a most likely user intent based on the outputs.