Distributed Speech Recognition for Voice Commands
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems face challenges in processing freely formulated voice commands, particularly when user-specific vocabulary is involved, due to limited local processing power and high error rates, and there are concerns about data protection and increased data transfer costs when uploading user data to servers.
Innovation Solution
A method that divides the task of recognizing and processing oral statements between two speech recognition devices, where a server-based system handles general vocabulary and a local device recognizes user-specific terms, keeping data local and reducing data transfer needs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If user data is uploaded to the server for recognition, then the recognition accuracy of user-specific vocabulary is improved, but data security and privacy protection deteriorate
Solution Approach 1:
The speech recognition task is segmented into two parts: server-based recognition for general vocabulary and local device recognition for user-specific vocabulary. This segmentation allows each component to focus on its strength while addressing the security concern by keeping sensitive user data local.
Solution Approach 2:
The system uses an intermediary approach where the server processes general speech recognition while a local speech recognition unit handles user-specific terms. This intermediary local unit acts as a bridge that protects user data privacy while still enabling accurate recognition of user-specific vocabulary.
2Measurement precision
If user data is uploaded to the server, then the recognition of user-specific vocabulary is improved, but data transfer volume and costs increase
Solution Approach 1:
The system extracts only the necessary user-specific vocabulary recognition task to the local device, while leaving general speech recognition on the server. This extraction minimizes data transfer by sending only general speech data to the server and keeping user-specific data local.
Solution Approach 2:
The recognition task is segmented between server and local device, with the local device handling user-specific terms. This segmentation reduces the volume of data that needs to be transferred to the server, as user-specific vocabulary remains local and only general speech content is transmitted.
3Object-affected harmful factors
If speech recognition is performed locally on the device, then data security is maintained, but the recognition accuracy of freely formulated utterances deteriorates
Solution Approach 1:
The system segments the recognition task by function: the server handles freely formulated utterances using its extensive vocabulary and processing power, while the local device handles user-specific vocabulary. This segmentation allows each component to operate at its optimal capability level.
Solution Approach 2:
The system implements a multi-functional architecture where the server provides universal speech recognition capabilities for general vocabulary, while the local device provides specialized recognition for user-specific terms. This multi-functionality ensures both high accuracy for freely formulated utterances and protection of user data privacy.
Data Source
Figure 1~2
Figure 3
Figure 4~7
AI summary
The invention relates to a method for generating a control command (B) from a verbal statement that contains unrestricted phrasing and user-specific terms, comprising the steps of: a) recording (1) of a voice command that comprises a multiplicity of words (W) as an audio data stream by a recording device (11); b) sending (2) of the audio data stream via a network (20) to a first voice recognition device (31); c) reception (5) of at least one data packet from the first voice recognition device (31), wherein the data packet contains information (I) concerning which words (W) in the audio data stream have not been recognised; d) at least partial recognition (7) of the words (W) that have not been recognised by the first voice recognition device (31) by a second voice recognition device (16) using at least one database; e) compilation (8) of the results from the first (31) and second (16) voice recognition devices to form a control command (B); and f) output of the control command (B).