Dual Speech Recognition Using Confidence Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition technologies in electronic devices face limitations in recognizing speech inputs due to capacity constraints, leading to inefficiencies and slower response rates, especially when relying solely on embedded models, and transmitting speech inputs to servers for recognition increases communication traffic and response times.
Innovation Solution
An electronic device employs a dual speech recognition approach by using a local speech recognition model for quick processing and transmitting uncertain inputs to a server for supplementary recognition, with confidence score thresholds determining the actions taken based on the recognition results from both models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If speech recognition is performed using an embedded speech recognition model in the electronic device, then the response speed is fast, but the recognition accuracy and reliability are limited due to device capacity constraints
Solution Approach 1:
The speech recognition system is segmented into two parts: an embedded speech recognition model in the electronic device for quick local processing, and a server-based speech recognition model for supplementary accurate recognition. This segmentation allows the system to leverage both the speed of local processing and the accuracy of cloud-based processing.
Solution Approach 2:
The embedded speech recognition model acts as an intermediary that pre-processes speech inputs locally and filters them before transmission to the server. This intermediary role reduces communication traffic while ensuring that only uncertain or complex inputs are sent for server processing, maintaining both speed and accuracy.
2Reliability
If speech inputs are transmitted to a server for speech recognition, then the recognition accuracy is improved, but the communication traffic increases and response time becomes slower
Solution Approach 1:
Instead of transmitting all speech inputs to the server, the system performs partial speech recognition locally using the embedded model and only transmits a subset of inputs that require supplementary recognition. This partial action approach reduces communication traffic and response time while maintaining recognition accuracy for complex inputs.
Solution Approach 2:
The embedded speech recognition model performs preliminary speech recognition before server processing. This preliminary action filters out simple, clear speech inputs that can be confidently processed locally, so only uncertain or complex inputs are transmitted to the server, reducing overall communication traffic and response time.
3Measurement precision
If the embedded speech recognition model is continuously updated with server recognition results, then the model accuracy is improved, but the device storage and processing load increase
Solution Approach 1:
The system extracts only the necessary updates and corrections from server recognition results, rather than storing complete server models. This extraction approach allows the embedded model to be continuously improved with server-level accuracy while maintaining a compact size suitable for device storage and processing capabilities.
Data Source
AI summary
An electronic device is provided. The electronic device includes a processor configured to perform automatic speech recognition (ASR) on a speech input by using a speech recognition model that is stored in a memory and a communication module configured to provide the speech input to a server and receive a speech instruction, which corresponds to the speech input, from the server. The electronic device may perform different operations according to a confidence score of a result of the ASR. Besides, it may be permissible to prepare other various embodiments speculated through the specification.