Speech Tendency Classification Using Multi-Acoustic Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems fail to effectively classify speech segments based on tendency categories, particularly emotional and non-emotional tendencies, due to a lack of comprehensive utilization of acoustic features.
Innovation Solution
A system that extracts acoustic features such as voiceprint, pitch, volume, and speaking rate information from speech segments to determine probabilities of tendency categories using machine learning models, including voiceprint-based, pitch-based, volume-based, and speaking rate-based models, which are combined to provide accurate classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple acoustic feature models are combined for speech tendency classification, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The system divides the speech analysis task into multiple independent acoustic feature extraction modules (voiceprint, pitch, volume, speaking rate), each processed by separate machine learning models. This segmentation allows each model to specialize in specific acoustic features, improving overall classification precision while maintaining modular system architecture that manages complexity.
Solution Approach 2:
The patent combines multiple machine learning models that process different acoustic features (voiceprint-based model, pitch-based model, volume-based model, speaking rate-based model) into an integrated speech tendency classification system. The outputs of these individual models are merged to produce the final classification result, leveraging complementary information from various acoustic dimensions to enhance measurement precision.
2Loss of information
If comprehensive acoustic feature extraction is performed, then information completeness is improved, but processing time increases
Solution Approach 1:
The system performs preliminary extraction of multiple acoustic features (voiceprint, pitch, volume, speaking rate) from speech segments before entering the machine learning classification stage. By preparing these acoustic feature representations in advance, the system ensures complete information is available for classification while enabling parallel processing of different features, which reduces overall processing time.
Solution Approach 2:
The patent implements continuous processing of acoustic features through multiple machine learning models in parallel, maintaining continuous useful action throughout the classification process. Instead of sequential analysis, the system processes voiceprint, pitch, volume, and speaking rate features simultaneously, ensuring no information is lost while minimizing processing time through concurrent operations.
Data Source
AI summary
Embodiments of the present disclosure relate to speech tendency classification. According to embodiments of the present disclosure, a method comprises extracting, from a speech segment, voiceprint information and at least one of volume information or speaking rate information; determining, based on the voiceprint information, first probability information indicating respective first probabilities of a plurality of tendency categories into which the speech segment is classified; determining, based on the at least one of the volume information or the speaking rate information, second probability information indicating respective second probabilities of the plurality of tendency categories into which the speech segment is classified; and determining, based at least in part on the first probability information and the second probability information, target probability information for the speech segment, the target probability information indicating respective target probabilities of the plurality of tendency categories into which the speech segment is classified.