Speech Break Detection Using EOS and EOU Boundary Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems struggle to accurately detect the ends of sentences and utterances in conversational AI systems, leading to suboptimal performance in processing text data, as they rely on capitalization, punctuation, and pauses, which are often unreliable indicators.
Innovation Solution
A system that utilizes both end of sentence (EOS) and end of utterance (EOU) detection models to identify breaks in speech, combining case and punctuation detection to improve accuracy in determining sentence and utterance endings, enhancing language model performance by processing specific text portions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional systems use capitalization, punctuation, and pause detection to identify sentence and utterance ends, then the system can detect breaks in speech, but the detection precision deteriorates because these indicators are unreliable (e.g., capitalized words in middle of sentences, periods in abbreviations, difficult pause detection)
Solution Approach 1:
The patent combines multiple detection approaches (capitalization detection, punctuation detection, pause detection, and duration-based detection) into a unified system. The break detection model integrates signals from all these sources, using a duration threshold parameter to filter false positives and improve overall precision by merging the strengths of each individual detection method.
Solution Approach 2:
The system dynamically adjusts the duration threshold parameter based on the detected break characteristics. By changing this parameter adaptively, the system can distinguish between actual sentence/utterance boundaries and false indicators, thereby improving detection precision while maintaining reliability across varying speech patterns.
2Productivity
If the system processes the entire utterance as a single text portion, then processing is simple, but language model performance deteriorates because multiple sentences within the utterance cannot be processed separately to improve precision
Solution Approach 1:
The patent segments the utterance into multiple sentence-level text portions based on detected sentence boundaries. This segmentation allows the language model to process each sentence separately, improving precision by enabling more targeted and context-aware processing of individual sentences while maintaining overall productivity through automated boundary detection.
Data Source
AI summary
In various examples, detecting breaks in speech for conversational AI systems and applications is described herein. Systems and methods are disclosed herein that use both end of sentence detection and end of utterance detection associated with words from text (e.g., tokens) to determine when to further process various portions of the text. For instance, one or more models may process text data associated with the text, where the text data may be generated using an automatic speech recognition (ARS) model based on audio data representing speech. Based at least on processing the text data, the model(s) may generate and/or output data representing first indicators that the words are associated with ends of sentences, second indicators that the words are associated with ends of utterances, and third indicators that the words are not associated with either ends of sentences or ends of utterances.