Voice Segmentation Using Two-Step Hysteresis Thresholding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer identification of voice from audio content is challenging due to the difficulty in distinguishing voice from other sounds, leading to inaccurate results.
Innovation Solution
A system that segments audio content into segments, determines feature vectors using Mel filter banks, processes them through classifiers like random forest or recurrent neural networks, and employs two-step hysteresis thresholding to identify voice segments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If voice identification is performed on audio content, then voice detection capability is achieved, but accuracy is insufficient due to difficulty in distinguishing voice from other sounds
Solution Approach 1:
The audio content is divided into multiple audio segments, and each segment is processed independently to determine whether it contains voice. This segmentation approach allows for more precise local analysis of voice characteristics in different time periods, improving overall detection accuracy while managing computational complexity.
Solution Approach 2:
The patent transforms the voice identification problem from direct audio signal analysis to feature vector space analysis. By extracting acoustic features and mapping them to a multidimensional feature space, the system can better distinguish voice from non-voice sounds through geometric relationships in the feature space, thereby improving measurement precision.
2Reliability
If feature vectors are processed through classifiers to identify voice segments, then voice identification capability is achieved, but classification stability is insufficient leading to inaccurate results
Solution Approach 1:
The system employs a feedback mechanism where classification results from adjacent audio segments are used to adjust and refine the classification of current segments. This contextual feedback from neighboring segments helps stabilize the classification process and reduces isolated misclassifications, thereby improving both reliability and precision.
Solution Approach 2:
The patent performs preliminary feature extraction and segmentation before classification. By pre-processing the audio content into standardized feature vectors and organizing them into segments with contextual information, the system prepares the data in advance for more stable and accurate classification, reducing the burden on the classifier and improving overall reliability.
Data Source
AI summary
Audio information defining audio content may be accessed. The audio content may have a duration. The audio content may be segmented into audio segments. Individual audio segments may correspond to a portion of the duration. Feature vectors of the audio segments may be determined. The feature vectors may be processed through a classifier. The classifier may output scores on whether the audio segments contain voice. One or more of the audio segments may be identified as containing voice based on the scores and a two-step hysteresis thresholding. Storage of the identification of the one or more of the audio segments as containing voice in one or more storage media may be effectuated.