A Convolutional Neural Network Smartphone App for Real-Time Voice Activity Detection.

Area of Science:

• Signal Processing
• Machine Learning
• Hearing Aid Technology

Background:

• Accurate voice activity detection (VAD) is crucial for effective noise reduction in hearing devices.
• Traditional VAD methods struggle with real-time performance and complex noise environments.
• Convolutional Neural Networks (CNNs) offer potential for improved VAD but face inference time challenges.

Purpose of the Study:

• To develop a real-time VAD smartphone application utilizing CNNs.
• To address and overcome the slow inference time limitations of CNNs for mobile applications.
• To enhance noise reduction capabilities in hearing devices through efficient VAD.

Main Methods:

• Implementation of a CNN-based VAD algorithm on a smartphone platform.
• Optimization techniques to accelerate CNN inference for real-time performance.
• Comparative analysis against a previous VAD app and established VAD algorithms.

Main Results:

• The developed CNN-based VAD smartphone app achieved real-time performance.
• The app successfully enabled noise estimation in noise-only segments of speech.
• Experimental results demonstrated superior performance compared to the previous app.

Conclusions:

• CNNs can be effectively implemented for real-time VAD on smartphones.
• The developed app provides a viable solution for adaptive noise reduction in hearing devices.
• This approach significantly improves upon existing VAD smartphone applications.