Voice Assistant Utterance Transformation via Language Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Electronic devices with voice assistant functions often struggle to process user utterances containing unrecognized words, leading to incomplete or incorrect interpretation of user intent.
Innovation Solution
The electronic device receives user utterance data, generates transformed utterance data using a language model and utterance-domain dataset, and provides a response based on this transformation, which includes candidate utterance data paired with various domains to handle unrecognized words.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the electronic device processes user utterances containing unrecognized words using traditional methods, then the processing speed is fast, but the accuracy of user intent determination deteriorates
Solution Approach 1:
The patent introduces an intermediary language model between the utterance recognition module and the intent determination module. This language model acts as a mediator that transforms unrecognized or ambiguous utterances into meaningful representations that can be processed for intent determination, thereby improving accuracy without requiring direct complex processing of raw unrecognized words
Solution Approach 2:
The patent performs preliminary processing of user utterances by generating candidate utterances and transforming them through a language model before intent determination. This preliminary action prepares the utterance data in advance, converting unrecognized words into processable forms that facilitate accurate intent determination while maintaining systematic efficiency
2Measurement precision
If the electronic device uses a language model to transform utterance data, then the interpretation accuracy improves, but the processing time increases
Solution Approach 1:
The patent applies partial action by selectively transforming only the portions of utterance data that contain unrecognized words or ambiguous meanings, rather than processing the entire utterance through the language model. This approach maintains interpretation accuracy for critical parts while reducing overall processing time
Solution Approach 2:
The patent segments the utterance processing into distinct stages: initial recognition, language model transformation of problematic segments, and intent determination. By dividing the processing pipeline and applying the language model only where necessary (to unrecognized words), the system achieves accurate interpretation while minimizing time loss
3Adaptability or versatility
If the electronic device generates transformed utterance data using candidate utterance data paired with multiple domains, then the versatility of handling different user inputs improves, but the computational complexity increases
Solution Approach 1:
The patent creates a universal language model that can handle multiple domains and types of unrecognized words through a single transformation framework. Instead of implementing separate processing systems for each domain, the language model provides multi-functional capability to transform various utterance types across different domains, thereby improving versatility without proportionally increasing computational complexity
Solution Approach 2:
The patent changes the parameters of utterance representation by transforming unrecognized words into standardized forms that work across multiple domains. By adjusting the representation parameters through language model transformation, the system achieves adaptability to handle different user inputs across various domains while maintaining consistent processing complexity
Data Source
AI summary
An electronic device includes a memory storing instructions; and a processor electrically connected to the memory and configured to execute the instructions to: receive user utterance data; obtain an utterance-domain data set including candidate utterance data that is based on the user utterance data; generate transformed utterance data associated with the user utterance data based on a language model and the utterance-domain data set; and provide a response corresponding to the user utterance data, based on the transformed utterance data. The utterance-domain data set may include at least one candidate utterance data paired with each of a plurality of domains. Each domain of the plurality of domains corresponds to a different operation or function.