Speech Endpointing Using User-Specific Pause Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional endpointers in natural language processing systems often inaccurately determine the start and end of a user's speech utterance based on pause duration, leading to incomplete or inaccurate processing of voice inputs.
Innovation Solution
A computing device assesses user experience by analyzing previous voice queries to adjust pause thresholds for determining speech endpoints, classifying users as novices or experts based on pause lengths, and generating endpoint signals accordingly to improve endpoint detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional endpointers use fixed pause duration thresholds to determine speech endpoints, then the system operates with simple processing logic, but the accuracy of speech endpoint detection deteriorates for users with different speaking patterns
Solution Approach 1:
The endpointer dynamically adjusts pause duration thresholds based on individual user characteristics. Instead of using fixed thresholds, the system adapts the pause duration criteria for determining speech endpoints according to each user's speaking patterns, thereby improving detection accuracy without requiring overly complex processing logic.
Solution Approach 2:
The system changes the parameter of pause duration threshold from a fixed value to a variable that depends on user-specific characteristics. By modifying this parameter based on observed user behavior patterns, the endpointer achieves higher accuracy in detecting speech endpoints for diverse users while maintaining relatively simple processing requirements.
2Productivity
If the endpointer segments voice input at long pauses, then processing speed is improved, but incomplete utterances are mistakenly processed leading to inaccurate results
Solution Approach 1:
The endpointer applies different pause duration criteria to different users based on their individual speaking characteristics. Rather than using a universal pause threshold, the system tailors the pause duration judgment to each user's local patterns, ensuring that segments are only processed when complete utterances are detected, thus maintaining both processing efficiency and accuracy.
3Adaptability or versatility
If a single pause threshold is used for all users, then the endpointer design is simple, but it fails to accommodate users with different speaking rhythms and pause patterns
Solution Approach 1:
The system performs preliminary classification of users into different categories (e.g., fast speakers, slow speakers, mid-speed speakers) based on their pause patterns. This preliminary action enables the endpointer to apply appropriate pause thresholds for each user category, improving adaptability to diverse speaking styles while keeping the overall system design relatively simple through categorical grouping.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, are described. In one aspect, a method includes accessing a collection of voice queries that were submitted by a particular user; determining a duration value based on durations of pauses between words of the voice queries in the collection of voice queries; receiving an utterance in which the particular user speaks one or more words, then pauses for less than the duration value, then speaks one or more other words; and submitting a voice query that includes a transcription of the one or more words that make up an initial portion of the utterance and the one or more other words, without first submitting a voice query that includes the transcription of the one or more words that make up the initial portion of the utterance without the one or more other words.