A hierarchical framework approach for voice activity detection and speech enhancement

Area of Science:

• Signal Processing
• Speech Technology
• Machine Learning

Background:

• Voice Activity Detection (VAD) is crucial for speech and speaker recognition systems.
• Existing VAD techniques struggle with performance in noisy environments.
• Speech enhancement is often required to improve the quality of degraded speech signals.

Purpose of the Study:

• To propose a hierarchical framework integrating VAD and speech enhancement.
• To improve the accuracy and robustness of VAD in noisy conditions.
• To evaluate the effectiveness of the proposed approach against established methods.

Main Methods:

• A hierarchical framework combining VAD and speech enhancement.
• Modified Wiener Filter (MWF) for noise reduction in speech enhancement.
• Feature selection and voting mechanism for reliable VAD decision-making.

Main Results:

• The proposed hierarchical framework demonstrates strong performance.
• The method is effective across various noisy conditions.
• Experimental results validate the approach using TIMIT and NOISEX-92 databases.

Conclusions:

• The integrated VAD and speech enhancement framework offers a robust solution.
• The proposed method shows significant improvements in noisy environments.
• This approach advances the field of robust speech processing.