Real-Time Speech Translation Using Buffered Sentence Splitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional sentence splitting methods struggle with determining split points in real-time, leading to challenges in providing accurate and timely language translation.
Innovation Solution
An information processing method that accumulates speech sound for a predetermined time or word count, detects sentence splits, and translates and outputs text or speech in a target language, with buffer replenishment and split point reevaluation to ensure accuracy and reduce waiting time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If sentences are continuously updated in real time, then translation responsiveness is improved, but split point determination becomes difficult
Solution Approach 1:
The patent accumulates a predetermined number of words or seconds of text in a buffer before attempting to detect split points. This preliminary accumulation ensures that sufficient text data is available for accurate sentence boundary detection, resolving the contradiction between real-time responsiveness and split point determination accuracy.
Solution Approach 2:
The patent segments the continuous text stream into discrete units by detecting sentence boundaries (split points) and processes each segment independently. This segmentation allows the system to maintain real-time processing while ensuring complete sentence context for accurate translation.
2Measurement precision
If split points are determined accurately, then translation precision is improved, but processing time increases
Solution Approach 1:
The system performs preliminary text accumulation in a buffer for a predetermined duration or word count before split point detection. This advance preparation ensures that when detection occurs, sufficient context is available for high-precision results, while the predetermined time limit prevents excessive processing delays.
Solution Approach 2:
The patent uses a predetermined threshold (first predetermined number of words or seconds) as a stopping criterion for text accumulation. This partial action approach ensures sufficient data for accurate detection without requiring complete processing of all available text, thereby reducing unnecessary processing time.
3Ease of manufacture
If text is accumulated for a fixed duration, then processing simplicity is improved, but translation accuracy may be compromised
Solution Approach 1:
The patent implements a dynamic accumulation strategy where the system monitors both time-based criteria (first predetermined number of seconds) and content-based criteria (first predetermined number of words). This dynamic approach allows the system to switch between time-limited and content-limited modes, maintaining processing simplicity while ensuring translation accuracy through flexible threshold adjustment.
Data Source
AI summary
Translation with high accuracy and with a short waiting time is provided. An information processing method by a terminal apparatus 1, the information processing method including: acquiring speech sound in a translation source language; recognizing the speech sound and generating text corresponding to the speech sound; accumulating a first predetermined number of seconds or a first predetermined number of words of the text in a buffer; detecting a split point of the text accumulated in the buffer, so as to detect a first sentence of the text; acquiring text in a translation target language that corresponds to the first sentence; and displaying the text in the translation target language and/or generating and outputting speech sound corresponding to the text in the translation target language.