Automatic switching between noise classification and speech enhancement for hearing aid devices

Area of Science:

• Audiology
• Signal Processing
• Machine Learning

Background:

• Hearing aid signal processing requires efficient methods for managing acoustic environments.
• Automatic switching between noise classification and speech enhancement is crucial for optimal hearing aid performance.
• Existing voice activity detectors (VADs) may have limitations in detection accuracy or processing speed.

Purpose of the Study:

• To develop and evaluate a novel voice activity detector (VAD) for hearing aid devices.
• To enable automatic and efficient switching between noise classification and speech enhancement modules.
• To compare the performance of the proposed VAD against established VAD methods.

Main Methods:

• A computationally efficient feature extractor was designed.
• A random forest classifier was implemented using established and novel signal features.
• The developed VAD was integrated into a hearing aid signal processing pipeline.
• Performance was evaluated against two popular VADs, focusing on detection rate and processing time.

Main Results:

• The developed VAD demonstrated superior performance compared to two popular VADs.
• The novel approach achieved a higher detection rate.
• The proposed VAD exhibited significantly faster processing times.
• The system effectively automated the switching between noise classification and speech enhancement.

Conclusions:

• The novel voice activity detector offers an improved solution for hearing aid signal processing.
• The VAD's efficiency in detection and speed makes it suitable for real-time hearing aid applications.
• This approach enhances the adaptability of hearing aids to varying acoustic environments.