Neural Network Voice Type Detection Using LPC Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies lack effective methods for real-time detection and classification of different voice types in audio data, which is crucial for applications like voice enhancement and emotion/user context detection.
Innovation Solution
A neural network-based system utilizing linear predictive coding (LPC) coefficients as input features to classify vocal efforts into distinct categories such as yelled, loud, regular, soft, and whispered voice types, enabling real-time identification and metadata generation for various applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional audio analysis methods are used, then the system is simple to implement, but real-time voice type detection accuracy is insufficient
Solution Approach 1:
The patent transforms the audio signal from time-domain to frequency-domain by extracting spectrogram features, and further transforms these features into a lower-dimensional embedding space using neural networks. This parameter transformation enables accurate voice type detection while maintaining real-time processing capability through efficient feature representation.
Solution Approach 2:
The patent replaces traditional mechanical audio analysis methods with a neural network-based system that uses deep learning models (including CNNs and transformers) to automatically learn and extract voice type characteristics from audio signals, achieving superior detection accuracy without manual feature engineering.
2Measurement precision
If complex neural network models are used for accurate detection, then voice type detection precision improves, but computational processing time increases
Solution Approach 1:
The patent divides the audio signal into overlapping frames (e.g., 25ms windows with 10ms hop), processing each frame independently through the neural network. This segmentation enables real-time processing of continuous audio streams while maintaining detection accuracy through frame-level analysis and aggregation.
Solution Approach 2:
The patent extracts only the most critical features from the full spectrogram (focusing on specific frequency ranges and temporal patterns relevant to voice type), rather than processing all possible audio characteristics. This selective feature extraction reduces computational load while preserving detection accuracy.
3Reliability
If comprehensive audio features are extracted for accurate classification, then detection reliability improves, but computational energy consumption increases
Solution Approach 1:
The patent extracts only the essential features needed for voice type detection (spectrogram characteristics, fundamental frequency, harmonics) from the full audio signal, discarding redundant information. This selective extraction maintains detection reliability while significantly reducing the computational energy required for processing.
4Productivity
If real-time processing is implemented, then system responsiveness improves, but measurement precision may deteriorate
Solution Approach 1:
The patent performs preliminary feature extraction and preprocessing on audio frames before they reach the main classification neural network. By pre-computing spectrograms and extracting key characteristics in advance, the system enables real-time processing without sacrificing detection precision during the main classification stage.
Data Source
AI summary
Methods, apparatus, systems, and articles of manufacture for real-time voice type detection in audio data are disclosed. An example non-transitory computer-readable medium disclosed herein includes instructions, which when executed, cause one or more processors to at least identify a first vocal effort of a first audio segment of first audio data and a second vocal effort of a second audio segment of the first audio data, train a neural network including training data, the training data including the first vocal effort, the first audio segment, the second audio segment, and the second vocal effort, and deploy the neural network, the neural network to distinguish between the first vocal effort and the second vocal effort.