Distributed Speech Recognition via Grammar Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated call center systems face delays and compromised quality due to the large search space created by rich grammars, leading to increased latency and decreased speech recognition accuracy, especially under high caller volume.
Innovation Solution
A distributed speech recognition system utilizing a centralized message server, a main speech recognizer, and multiple secondary recognizers, where the message server initiates sessions with each secondary recognizer, assigning non-overlapping grammar sections for parallel processing, and employing an n-best algorithm to construct a new grammar for efficient recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a rich grammar is used to improve speech recognition accuracy, then recognition quality is improved, but search space increases and latency increases exponentially
Solution Approach 1:
The patent divides the rich grammar into multiple non-overlapping sections and assigns each section to a different secondary speech recognizer. This segmentation allows parallel processing of speech recognition tasks across multiple recognizers, reducing the search space for each individual recognizer while maintaining comprehensive coverage of the entire grammar. The main speech recognizer then integrates results from all secondary recognizers to achieve accurate speech recognition with reduced latency.
2Measurement precision
If a rich grammar is used to improve speech recognition accuracy, then recognition quality is improved, but device complexity increases
Solution Approach 1:
The system segments the complex speech recognition task into multiple simpler sub-tasks, each handled by a secondary recognizer with a portion of the grammar. This distribution of complexity across multiple independent components makes the overall system more manageable and scalable, while the main recognizer coordinates these components to achieve accurate recognition.
Solution Approach 2:
The main speech recognizer acts as an intermediary that coordinates between the secondary recognizers and the final speech recognition output. It receives results from multiple secondary recognizers, integrates them using the n-best algorithm, and produces the final accurate recognition result, thereby managing system complexity through hierarchical organization.
3Extent of automation
If conventional automated systems are used to reduce reliance on human agents, then automation is improved, but response delays increase under high caller volume
Solution Approach 1:
The patent implements a segmented automated speech recognition system with multiple secondary recognizers processing speech simultaneously in parallel. This parallel processing capability allows the automated system to handle high caller volumes efficiently without increasing response delays, maintaining high automation levels while improving response time under load.
Data Source
AI summary
A computer-implemented system and method for performing distributed speech recognition is provided. Audio data is collected. A main grammar and secondary grammars are simultaneously provided for the audio data. Each secondary grammar includes an independent grammar. Speech recognition is simultaneously performed on the audio data using each secondary grammar. A new grammar is constructed for the audio data based on the main grammar template using results of the speech recognition. Further speech recognition is performed on the audio data using the new grammar.