Speech Recognition Close Call Records for Memory Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems in handheld devices face challenges due to limited processing power and memory, leading to incorrect decoding of utterances and inefficiencies in generating alternative versions of recognized transcripts.
Innovation Solution
A method that generates a list of close call records during speech recognition, comparing competing partial hypotheses to create alternative transcripts, which reduces memory requirements and maintains effectiveness regardless of utterance length, by using a preset maximum number of close call records and scoring differences to construct the choice list.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing speech recognition techniques are used to generate alternative utterances, then the system can provide choice lists for error correction, but the memory requirements increase significantly and processing power demands are excessive for handheld devices
Solution Approach 1:
The patent extracts only the essential information needed for generating alternative utterances - specifically close call records with competing hypotheses and their score differences - while discarding unnecessary data. This selective extraction reduces memory requirements from storing complete alternative lists to storing only the minimal close call records needed for hypothesis generation.
Solution Approach 2:
The patent segments the speech recognition process into distinct phases: generating close call records during recognition, storing them in a limited buffer, and then generating alternative utterances by processing these segmented records. This segmentation allows the system to handle data in manageable portions rather than requiring large memory buffers for the entire utterance history.
2Measurement precision
If larger buffers are used to store more close call records, then the choice list quality improves, but the memory requirements and processing burden increase
Solution Approach 1:
The patent changes the parameter of buffer size from a fixed large value to a configurable parameter that can be optimized for handheld devices. By adjusting the maximum number of close call records stored, the system achieves acceptable choice list quality with reduced processing complexity and lower memory requirements suitable for portable platforms.
3Measurement precision
If the system processes longer utterances, then the recognition accuracy improves, but the memory requirements increase linearly with utterance length
Solution Approach 1:
The patent performs preliminary action by generating and storing close call records during the speech recognition process itself, rather than waiting to process entire utterances afterward. This allows the system to maintain recognition accuracy for longer utterances while using a fixed, manageable buffer size for storing only the critical close call moments, not the entire utterance history.
Data Source
AI summary
A method of constructing a list of alternate transcripts from a recognized transcript includes generating a list of close call records, matching partial sub-histories from the recognized transcript with one of the history pairs stored in each of the records, and substituting the other of the history pairs for the partial sub-history of the recognized transcript. A close call record is generated each time a pair of partial hypotheses attempt to seed a common word. Each close call record includes history information and scoring information associated with a particular pair of partial hypotheses seeding a common word. Alternate transcripts are constructed by substituting close call histories for partial histories of the recognized transcripts, and also by substituting close call histories for partial histories of other alternate transcript.