Audio Signal Segmentation Using Breath Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio notetaking systems struggle to accurately segment voice recordings in noisy environments, leading to misidentification of spoken phrases due to difficulty in detecting pauses, especially when speakers are distant or speak quickly/quietly.
Innovation Solution
A method that identifies spoken phrases by detecting voice activity and inactivity, and if a segment exceeds a threshold duration, it searches for the most likely breath location to split the segment into sub-segments, ensuring accurate segmentation by assuming a breath occurred within the long segment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pause detection is used to segment audio into spoken phrases, then segmentation can be performed based on natural speech boundaries, but in noisy environments or when speakers are distant, pause detection becomes unreliable and segmentation accuracy deteriorates
Solution Approach 1:
The system changes the segmentation parameter from pause detection to breath location detection. By analyzing breath sounds (which have distinct acoustic characteristics) within long segments, the system identifies more reliable boundaries for spoken phrases, thereby maintaining segmentation accuracy in noisy environments where pause detection fails.
Solution Approach 2:
The system introduces breath location as an intermediary marker to determine phrase boundaries. Instead of directly detecting pauses (which are unreliable in noise), the system uses breath sounds as intermediate indicators that precede or follow pauses, providing a more robust method for identifying phrase boundaries in challenging acoustic conditions.
2Device complexity
If long segments are kept intact, then fewer segmentation operations are needed, but the segments may not align with actual speech phrases reducing usability
Solution Approach 1:
The system applies segmentation by dividing long audio segments into smaller sub-segments based on detected breath locations. This creates multiple manageable units that align with actual speech phrases, improving usability for live notetaking while maintaining a relatively simple overall process through automated breath detection.
3Measurement precision
If breath location detection is performed on all segments, then segmentation accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The system performs preliminary segmentation based on simpler criteria first, identifying long segments that require further processing. Breath location detection is then applied only to these identified long segments rather than all segments, reducing unnecessary computational effort while maintaining accuracy where it matters most.
Solution Approach 2:
The system applies the more computationally intensive breath detection method partially - only to segments that exceed a certain duration threshold. This partial application of the rigorous method optimizes the balance between accuracy and processing efficiency by avoiding excessive computation on already-short segments.
Data Source
AI summary
The present disclosure relates to methods, computer programs, and computer-readable media for processing a voice audio signal. A method includes receiving, at an electronic device, a voice audio signal, identifying spoken phrases within the voice audio signal based on the detection of voice activity or inactivity, dividing the voice audio signal into a plurality of segments based on the identified spoken phrases, and in accordance with a determination that a selected segment of the plurality of segments has a duration, Tseg, longer than a threshold duration, Tthresh, identifying a most likely location of a breath in the audio associated with the selected segment and dividing the selected segment into sub-segments based on the identified most likely location of a breath.