Offline Voice Recognition via Compressed Acoustic Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voice recognition technologies rely on network connectivity, limiting their functionality and usability, especially in mobile devices where resource constraints hinder the execution of complex acoustic models.
Innovation Solution
Implementing a method and device for offline voice recognition by collecting and decoding voice information using pre-compressed acoustic and language models, which enables voice recognition without network dependency and optimizes data processing through data compression and parallel computation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If voice recognition is performed online using cloud servers, then recognition accuracy is maintained, but network dependency increases and offline functionality is lost
Solution Approach 1:
The patent divides the voice recognition system into separate functional modules: acoustic model, language model, and decoding module. These segmented components can be selectively deployed and executed on mobile devices, enabling offline functionality while maintaining recognition accuracy through distributed computation
Solution Approach 2:
The patent introduces a data compression technology as an intermediary between the complex acoustic models and mobile device resources. This intermediary enables the transmission and execution of compressed models on resource-constrained devices, bridging the gap between cloud-based accuracy and offline capability
2Adaptability or versatility
If complex acoustic models are deployed on mobile devices, then offline voice recognition is achieved, but device resources are exceeded and processing becomes infeasible
Solution Approach 1:
The patent applies data compression to transform the acoustic model parameters into a compressed format. This parameter transformation reduces the model size by approximately 95% while preserving recognition accuracy, making complex models executable on mobile devices with limited resources
Solution Approach 2:
The patent performs data compression on acoustic models in advance before deployment to mobile devices. This preliminary processing reduces model complexity beforehand, enabling offline voice recognition without requiring significant device resources during actual recognition operations
3Measurement precision
If uncompressed acoustic models are used, then model accuracy is maintained, but decoding time increases significantly
Solution Approach 1:
The patent transforms model parameters through compression, creating a compact representation that maintains accuracy while enabling faster processing. The compressed parameters are optimized for rapid computation on mobile devices, reducing decoding time without sacrificing recognition precision
Data Source
AI summary
A method for recognizing a voice and a device for recognizing a voice are provided. The method includes: collecting voice information input by a user; extracting characteristics from the voice information to obtain characteristic information; decoding the characteristic information according to an acoustic model and a language model obtained in advance to obtain recognized voice information, wherein the acoustic model is obtained by data compression in advance.