ASR Data Tiling for Resource-Constrained GPS Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional automated speech recognition (ASR) techniques consume significant processing resources and become complicated when dealing with large amounts of data, such as in devices with position-determining functionality like GPS, which have limited resources.
Innovation Solution
Implementing ASR data tiling by dividing data into geographic partitions and using common tiles to reduce data redundancy and improve resource efficiency, allowing for faster and more accurate speech recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional ASR techniques are implemented to handle large amounts of data in position-determining devices, then speech recognition capability is provided, but processing resources are significantly consumed
Solution Approach 1:
The patent segments the large ASR data into multiple tiles based on geographic regions. Each tile contains speech recognition data for a specific geographic area. The system loads only the tiles corresponding to the current or anticipated geographic region, rather than loading all ASR data into memory. This segmentation reduces the amount of data that needs to be processed and stored in memory, thereby reducing processing resource consumption while maintaining speech recognition capability for location-specific queries.
2Adaptability or versatility
If traditional ASR techniques are implemented with large amounts of data, then comprehensive speech recognition is achieved, but device complexity increases
Solution Approach 1:
The patent divides the comprehensive ASR data into geographic tiles, organizing the complexity into a manageable structure. Each tile is self-contained and represents a specific geographic region. The system manages complexity by tracking which tiles are needed based on GPS location and loading only those tiles into memory. This segmentation transforms the complexity of managing large datasets into a simpler geographic-based organization system that is easier to manage and access.
Solution Approach 2:
The patent performs preliminary actions by pre-organizing ASR data into geographic tiles before the device needs to use them. The data is pre-segmented and stored in a structured format that can be quickly loaded based on location. This preliminary organization eliminates the need for complex real-time data processing and management during operation, reducing device complexity while maintaining comprehensive speech recognition coverage.
3Measurement precision
If all ASR data is loaded into memory for processing, then speech recognition accuracy is improved, but memory resources are exhausted in portable devices
Solution Approach 1:
The patent segments ASR data into geographic tiles that can be selectively loaded into memory based on the device's current location. Only the tiles corresponding to the current or anticipated geographic region are loaded into memory, rather than loading all ASR data. This approach maintains speech recognition accuracy for location-specific queries by having the relevant data in memory, while conserving memory resources by not loading unnecessary data from other geographic regions.
Solution Approach 2:
The patent applies local quality by optimizing memory usage for the specific geographic context. The system loads high-quality, location-specific ASR data into memory for the current region, while storing less frequently accessed data from other regions on external storage. This ensures that the most relevant data is available in memory for accurate speech recognition, while managing overall memory resources efficiently according to local geographic needs.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Techniques are described related to tiles of automated speech recognition data. In an implementation, automated speech recognition (ASR) data is obtained. The ASR data is divided into a plurality of tiles based on an approximate amount of data to be included in each tile. Each of the tiles is a partition of the ASR data.