Hybrid Speech Recognition System for Latency and Accuracy Trade-off
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice interface systems using automated speech recognition (ASR) face delays and quality degradation due to the need to transmit audio signals for processing, which can be cumbersome for devices with limited processing power and memory, and often rely solely on either local or remote ASR systems without effectively combining their strengths.
Innovation Solution
A hybrid speech recognition system that combines local and network-based ASR systems, using local recognition for immediate actions and network recognition for verification, with parallel processing and a priori weighting to improve accuracy and reduce latency, while also considering user-specific knowledge and network resource management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If audio is transmitted to network-based recognizer for processing, then recognition accuracy is improved, but transmission time and quality degradation occur
Solution Approach 1:
The patent segments the speech recognition task into two parts: local ASR handles immediate recognition needs while network-based ASR provides verification for accuracy-critical tasks. This segmentation allows the system to achieve both speed (from local processing) and accuracy (from network verification) without the trade-off present in single-system approaches.
Solution Approach 2:
The patent merges local and network-based ASR systems into a hybrid architecture where both systems process the same audio input and their results are combined. The local ASR provides immediate results while the network-based ASR provides verification, creating a system that achieves both speed and accuracy simultaneously.
2Loss of time
If local ASR is used for immediate recognition, then response time is reduced, but recognition accuracy deteriorates
Solution Approach 1:
The patent applies partial action by having the local ASR handle only the immediate recognition task while deferring verification to the network-based ASR. This allows the system to provide quick responses based on local processing while maintaining accuracy through subsequent network verification, rather than requiring the local system to handle all processing alone.
Solution Approach 2:
The network-based ASR results serve as feedback to verify and correct the local ASR results. The system uses the network-based recognition results to feedback-correct any errors in the local ASR output, thereby maintaining high accuracy while preserving the speed advantage of local processing.
3Productivity
If audio is compressed for network transmission, then transmission efficiency is improved, but audio quality degrades
Solution Approach 1:
The patent segments the audio processing into two paths: local processing uses uncompressed audio for immediate recognition, while network transmission uses compressed audio for verification. This segmentation allows the system to optimize each path for its specific requirements without compromising overall system performance.
Data Source
AI summary
Techniques for combining the results of multiple recognizers in a distributed speech recognition architecture. Speech data input to a client device is encoded and processed both locally and remotely by different recognizers configured to be proficient at different speech recognition tasks. The client/server architecture is configurable to enable network providers to specify a policy directed to a trade-off between reducing recognition latency perceived by a user and usage of network resources. The results of the local and remote speech recognition engines are combined based, at least in part, on logic stored by one or more components of the client/server architecture.