Speech Processing Apparatus Morphological Boundary Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech processing technologies face challenges in accurately dividing human utterances into appropriate processing pieces due to varying pause lengths among speakers, leading to either prolonged processing times or choppy segments, which degrade the simultaneity and accuracy of subsequent processing.
Innovation Solution
A simultaneous speech processing apparatus comprising an acquisition unit, speech recognition unit, detection unit, and output unit that generates decided and candidate character strings using speech recognition, and detects processing pieces by analyzing combined character strings with morphological boundaries, employing methods like conditioning random fields and machine learning to optimize processing piece detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If speech is divided into processing pieces based on pause length, then processing can be performed in parallel, but the accuracy of processing piece division deteriorates due to varying pause lengths among speakers
Solution Approach 1:
The patent segments the speech recognition result into multiple processing pieces by detecting morphological boundaries (such as sentence boundaries, clause boundaries, or phrase boundaries) rather than relying solely on pause length. This allows parallel processing while maintaining accuracy, as the segmentation is based on linguistic structure rather than temporal characteristics that vary across speakers.
Solution Approach 2:
The patent changes the parameter used for dividing speech from temporal parameters (pause length) to linguistic parameters (morphological boundaries). By detecting boundaries based on morphological analysis of the recognized text, the system achieves both high processing speed through parallelization and high division accuracy by using speaker-independent linguistic features.
2Ease of manufacture
If a fixed pause length threshold is used for dividing speech, then the division process is simple, but the simultaneity of processing deteriorates due to varying pause lengths
Solution Approach 1:
The patent replaces the fixed pause length threshold with a dynamic boundary detection mechanism that identifies morphological boundaries in the speech recognition result. This dynamic approach adapts to the actual linguistic structure of the speech, enabling faster processing by directly identifying meaningful units without waiting for fixed-duration pauses, thus reducing processing time while maintaining simplicity through automated boundary detection.
3Loss of time
If speech is divided into many small processing pieces, then processing can start sooner, but the accuracy of subsequent processing deteriorates due to choppy segments
Solution Approach 1:
The patent segments speech into processing pieces based on morphological boundaries (sentence boundaries, clause boundaries, phrase boundaries) rather than arbitrary time intervals. This ensures that each processing piece is a meaningful linguistic unit, maintaining high processing accuracy while enabling early start of processing. The segmentation is driven by linguistic structure rather than temporal constraints, preventing choppy segments.
Solution Approach 2:
The patent performs preliminary morphological analysis of the speech recognition result to identify potential processing piece boundaries before actual processing begins. This preliminary action allows the system to prepare appropriate processing pieces in advance based on linguistic structure, ensuring both early processing start and high accuracy by pre-identifying meaningful units rather than waiting for temporal cues.
Data Source
AI summary
According to one embodiment, a simultaneous speech processing apparatus includes an acquisition unit, a speech recognition unit, a detection unit and an output unit. The acquisition unit acquires a speech signal. The speech recognition unit generates a decided character string and at least one candidate character string. The detection unit detects a first character string as a processing piece character string if the first character string included in the decided character string exists commonly in one or more combined character strings on dividing the one or more combined character strings by a boundary indicating a morphological position serving as a start position of a processing piece in natural language processing. The output unit outputs the processing piece character string.