Client-Server Semantic Parsing for Speech Transcription Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems face inaccuracies in semantic parsing due to limited access to user-specific data, such as contact lists and calendars, which are not available to server-based semantic parsers, leading to incomplete or inaccurate transcription evaluations.
Innovation Solution
Implementing a dual semantic parsing approach where the client device receives candidate transcriptions from a server, performs additional parsing using locally accessible data, and updates semantic classifications and scores, allowing for more accurate evaluation and action selection based on user-specific information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If server-based semantic parsing is used to evaluate candidate transcriptions, then computational load on client device is reduced, but parsing accuracy deteriorates due to lack of access to user-specific data
Solution Approach 1:
The semantic parsing process is divided into two segments: server-based initial semantic parsing that handles general transcription evaluation, and client-based additional semantic parsing that leverages local user-specific data. This segmentation allows each component to perform its specialized function, with the server reducing computational load and the client improving accuracy through access to contact lists, calendars, and other personal information.
2Measurement precision
If client device performs additional semantic parsing using local data, then parsing accuracy is improved, but device complexity increases
Solution Approach 1:
The server performs preliminary semantic parsing to generate initial semantic information and candidate transcriptions before the client device receives them. This preliminary action reduces the amount of work the client device must perform, as it only needs to conduct additional parsing on already-filtered candidates rather than processing raw transcriptions from scratch, thereby managing device complexity while improving accuracy.
3Measurement precision
If multiple candidate transcriptions are evaluated semantically, then transcription accuracy is improved, but processing time increases
Solution Approach 1:
Instead of performing exhaustive semantic parsing on all possible transcriptions, the system applies partial action by evaluating only the top N candidate transcriptions (n-best hypotheses) that are most likely to be correct based on speech recognition confidence scores. This approach achieves sufficient transcription accuracy by focusing computational resources on the most promising candidates rather than processing every possibility.
Data Source
AI summary
In some implementations, audio data for an utterance is provided over a network. At a client device and over the network, information is received that indicates candidate transcriptions for the utterance and semantic information for the candidate transcriptions. A semantic parser is used at the client device to evaluate each of at least a plurality of the candidate transcriptions. One of the candidate transcriptions is selected based on at least the received semantic information and the output of the semantic parser for the plurality of candidate transcriptions that are evaluated.