Local Dictation Database for Offline Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Organizations face challenges with software installation and maintenance on in-house computer systems, including time-consuming processes, resource-intensive updates, compatibility issues, and the need for virus protection, leading to outdated software and compatibility problems across different systems.
Innovation Solution
Implementing an on-demand database service that allows mobile devices to interact with a remote server for speech recognition, enabling the creation of a local dictation database that maintains dictation functionality offline and updates the database with new terms received from the server, reducing the need for continuous internet connection and processing power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If speech recognition processing is performed entirely on a remote server, then recognition accuracy is improved, but network dependency increases and offline functionality is lost
Solution Approach 1:
The speech recognition system is segmented into two parts: a comprehensive recognition model on the remote server and a smaller local vocabulary database on the mobile device. This segmentation allows the device to handle common terms offline while maintaining the option to use the full server-based system when connected, thus resolving the contradiction between offline capability and recognition accuracy.
Solution Approach 2:
A local vocabulary database is pre-loaded on the mobile device with commonly used terms and phrases. This preliminary action enables the device to perform speech recognition offline for these pre-loaded terms, while still allowing accurate recognition of new or less common terms when connected to the server, thereby balancing offline functionality with recognition accuracy.
2Reliability
If a local database is maintained on mobile devices, then offline dictation functionality is improved, but device storage requirements increase
Solution Approach 1:
Instead of storing the complete speech recognition vocabulary on the mobile device, only a partial set of commonly used terms is stored locally. This partial action provides sufficient offline functionality for everyday use while minimizing storage requirements, and allows the system to supplement with server-based recognition when needed.
3Measurement precision
If speech recognition models are frequently updated, then recognition accuracy is improved, but maintenance complexity increases
Solution Approach 1:
The complex speech recognition model is extracted from the mobile device and placed on the remote server, where updates can be managed centrally. The device only maintains a simpler local vocabulary database, which requires minimal updates. This extraction resolves the maintenance complexity issue while preserving the ability to update recognition accuracy on the server side.
4Use of energy by moving object
If all speech processing is done on remote servers, then processing power requirements at device are reduced, but network dependency increases
Solution Approach 1:
The system implements local quality by equipping the mobile device with a small but functional local vocabulary database for offline speech recognition. This local capability provides network independence for common terms, while the device can still leverage server-based processing for more complex recognition tasks when connected, thus balancing energy savings with adaptability.
Data Source
AI summary
Disclosed are methods, apparatus, systems, and computer-readable storage media for selectively interacting with a server to build a local dictation database for speech recognition at a device. In some implementations, a computing device receives an audio sample. The computing device may determine that the received audio sample does not match any of one or more existing audio samples stored in the local dictation database of the computing device. The received audio sample may be transmitted to a remote server for detection of one or more words indicated by the received audio sample. The computing device may receive data identifying the one or more words, and update the local dictation database to store the received audio sample in association with the one or more words.