Environmental Noise Classification with Inception-Dense Blocks for Hearing Aids.

Area of Science:

• Audiology and Signal Processing
• Machine Learning for Hearing Healthcare

Background:

• Environmental noise estimation and classification are crucial for modern hearing aids.
• Existing noise classifiers often suffer from high computational demands and sensitivity to input duration.

Purpose of the Study:

• To propose a novel, computationally efficient model architecture for environmental noise classification in hearing aids.
• To evaluate the impact of different audio segment lengths on classification performance.
• To reduce computational complexity and inference time without sacrificing accuracy.

Main Methods:

• Developed a model utilizing log-scaled mel-spectrograms as input features to minimize operations and parameters.
• Experimented with three distinct audio segment time lengths.
• Evaluated performance based on classification accuracy, computational complexity, trainable parameters, and inference time.

Main Results:

• The proposed model demonstrated superior performance over existing methods on the UrbanSound8k and HANS datasets.
• Achieved reduced model complexity and faster inference times.
• Maintained high classification accuracy comparable to or better than other models.

Conclusions:

• The developed noise classification model offers a significant reduction in computational complexity and inference time for hearing aid applications.
• This approach provides an efficient solution for environmental noise classification, enhancing hearing aid functionality without performance compromise.