Circular Buffer Speech Endpoint Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems often produce incomplete or inaccurate results due to imperfect synchronization between user speech and system commands, as well as poor endpointing in noisy environments, leading to inefficiencies and increased computational resources.
Innovation Solution
A method that continuously records audio streams and uses Hidden Markov Models (HMMs) to identify and process augmented audio signals, ensuring complete speech recognition by retrieving missed portions and improving endpoint detection in real-time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional endpointing methods use short-time energy or spectral energy features, then the system can determine speech endpoints, but the reliability decreases in noisy real-world situations
Solution Approach 1:
The patent introduces an intermediary mechanism that continuously records audio streams in a circular buffer and retrieves augmented audio signals surrounding the user command. This intermediary buffer allows the system to access complete speech signals even when endpointing is triggered during noisy periods, effectively mediating between the noisy environment and the speech recognition processing.
Solution Approach 2:
The system performs preliminary action by continuously recording audio streams before the actual speech recognition processing occurs. The circular buffer stores audio frames in advance, so when endpointing is needed, the system already has access to the complete audio signal including portions before and after the command, ensuring reliable endpoint detection even in noisy conditions.
2Measurement precision
If the speech recognition system waits for complete speech signal synchronization, then processing accuracy improves, but response time increases
Solution Approach 1:
The system performs preliminary action by continuously recording audio streams in the circular buffer before the user provides the command. When the user commands speech recognition, the system immediately processes the augmented audio signal that was already recorded, rather than waiting for complete synchronization. This allows the system to maintain high accuracy by having the complete speech signal ready while minimizing response time.
Solution Approach 2:
The system uses feedback from the circular buffer to retrieve augmented audio signals that contain the complete speech information. The feedback mechanism allows the system to adjust the retrieval window based on the actual speech duration and command timing, ensuring that the complete speech signal is always available for processing regardless of synchronization delays.
3Productivity
If the system retrieves augmented audio signals using fixed time windows, then processing speed is maintained, but completeness of speech signals is lost
Solution Approach 1:
The system dynamically adjusts the audio signal retrieval window based on the actual speech characteristics and command timing. Rather than using fixed time windows, the system calculates the appropriate retrieval duration based on the detected speech endpoints and command offset, ensuring that the complete speech signal is always retrieved while maintaining efficient processing speed.
Solution Approach 2:
The system uses feedback from speech endpoint detection to dynamically adjust the audio retrieval parameters. When speech endpoints are detected, the system uses this feedback information to calculate the optimal retrieval window that captures the complete speech signal, ensuring both completeness and processing efficiency.
Data Source
AI summary
The present invention relates to a method and apparatus for obtaining complete speech signals for speech recognition applications. In one embodiment, the method continuously records an audio stream comprising a sequence of frames to a circular buffer. When a user command to commence or terminate speech recognition is received, the method obtains a number of frames of the audio stream occurring before or after the user command in order to identify an augmented audio signal for speech recognition processing. In further embodiments, the method analyzes the augmented audio signal in order to locate starting and ending speech endpoints that bound at least a portion of speech to be processed for recognition. At least one of the speech endpoints is located using a Hidden Markov Model.