Speech Detection Using Cepstral Coefficients and SVM
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer identification of speech from audio content is difficult and often inaccurate due to the challenge of distinguishing speech from other sounds.
Innovation Solution
The system segments audio content into portions, extracts Mel frequency spectral power features, Mel frequency cepstral coefficient features, and energy features, and uses a support vector machine to process these features to identify speech, with filtering and threshold comparison to determine the presence of speech in audio segments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computer identification of speech from audio content is attempted, then speech detection capability is provided, but accuracy is poor due to difficulty distinguishing speech from other sounds
Solution Approach 1:
The audio content is divided into multiple audio segments, allowing the system to analyze speech characteristics in smaller, manageable portions. This segmentation enables more precise detection by comparing features within each segment against speech patterns, thereby improving accuracy while managing the complexity of distinguishing speech from other sounds.
Solution Approach 2:
The system extracts multiple types of features (spectral power features, cepstral coefficient features, energy features) from each audio segment and transforms them into feature vectors. By changing and analyzing multiple parameters simultaneously, the support vector machine can more accurately distinguish speech from non-speech sounds, resolving the contradiction between detection difficulty and accuracy.
2Measurement precision
If multiple feature extraction methods are used to improve speech identification accuracy, then detection precision improves, but system complexity increases
Solution Approach 1:
The system divides the complex task of speech identification into separate feature extraction stages: spectral power feature extraction, cepstral coefficient feature extraction, and energy feature extraction. Each stage handles a specific aspect of audio analysis, which reduces overall system complexity while maintaining high accuracy through the combination of multiple specialized features.
Solution Approach 2:
The support vector machine acts as an intermediary that receives feature vectors from multiple extraction processes and integrates them into a unified speech identification decision. This intermediary component manages the complexity of combining multiple feature types while preserving the accuracy benefits of using diverse feature extraction methods.
Data Source
AI summary
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. Mel frequency spectral power features, Mel frequency cepstral coefficient features, and energy features of the audio segments may be determined. Feature vectors of the audio segments may be determined based on the Mel frequency spectral power features, the Mel frequency cepstral coefficient features, and the energy features. The feature vectors may be processed through a support vector machine. The support vector machine may output predictions on whether the audio segments contain speech. One or more of the audio segments may be identified as containing speech based on filtering the predictions and comparing the filtered predictions to a threshold. Storage of the identification of the one or more of the audio segments as containing speech in one or more storage media may be effectuated.