Voice Activity Detection Using Zero Crossing and Pulse Density
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voice activity detection (VAD) systems in smart speakers and voice-controlled devices are inefficient in distinguishing noise from human speech, leading to increased power consumption and potential false positives.
Innovation Solution
The implementation of a VAD system that utilizes a combination of band-pass filtering, zero crossing detection, pulse density analysis, and statistical threshold adjustment to accurately differentiate between speech and noise, dynamically adjusting parameters based on ambient noise and signal characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the device continuously listens for instructions to ensure voice control functionality, then the reliability of voice instruction detection is improved, but the power consumption increases
Solution Approach 1:
The system performs preliminary voice activity detection using low-power zero-crossing detection before engaging full speech processing. This preliminary action filters out non-speech segments early, allowing the device to maintain reliability while reducing overall power consumption by avoiding continuous full-processing mode
Solution Approach 2:
The patent replaces complex continuous speech processing with a simplified zero-crossing detection mechanism for initial filtering. This substitution uses a less computationally intensive method to achieve the same goal of identifying potential speech segments, thereby reducing power consumption while maintaining detection reliability
2Measurement precision
If the device processes all audio signals to ensure no speech is missed, then the detection accuracy is improved, but the false positive rate increases due to noise
Solution Approach 1:
The audio signal processing is segmented into multiple stages: initial zero-crossing detection, followed by energy detection, and finally full speech processing only for segments that pass earlier filters. This segmentation allows the system to maintain high detection accuracy while reducing false positives by applying progressively more sophisticated analysis only where needed
Solution Approach 2:
Different processing qualities are applied to different segments of the audio signal. Quiet segments undergo rigorous multi-stage analysis to avoid missing speech, while noisy segments are filtered more aggressively. This local quality adjustment optimizes the balance between detection accuracy and false positive reduction based on the specific characteristics of each audio segment
3Device complexity
If simple noise thresholding is used to reduce processing complexity, then the device complexity is reduced, but the measurement precision of speech detection deteriorates
Solution Approach 1:
The detection system dynamically adjusts its complexity based on signal characteristics. Zero-crossing detection provides a simple initial filter, but the system transitions to more precise energy-based and spectral analysis when speech is detected. This dynamic adjustment maintains low average complexity while ensuring high precision when needed
Solution Approach 2:
The system employs periodic analysis with increasing sophistication: first zero-crossing rate analysis, then energy detection in periodic windows, and finally full spectral analysis only when needed. This periodic progression through analysis stages maintains simplicity for most of the time while ensuring precision when speech is present
Data Source
AI summary
A first VAD system outputs a pulse stream for zero crossings in an audio signal. The pulse density of the pulse stream is evaluated to identify speech. The audio signal may have noise added to it before evaluating zero crossings. A second VAD system rectifies each audio signal sample and processes each rectified sample by updating a first statistic and evaluating the rectified sample per a first threshold condition that is a function of the first statistic. Rectified samples meeting the first threshold condition may be used to update a second statistic and the rectified sample evaluated per a second threshold condition that is a function of the second statistic. Rectified samples meeting the second threshold condition may be used to update a third statistic. The audio signal sample may be selected as speech if the second statistic is less than a downscaled third statistic.