Development of a voice activity controlled noise canceller.

Area of Science:

• Signal Processing
• Acoustics
• Speech Technology

Background:

• Adaptive noise cancellation (ANC) systems require effective voice activity detection (VAD) to optimize performance.
• Traditional VAD methods can be suboptimal when speech signals are present in the reference input during adaptation.
• Efficient noise reduction is crucial in various audio processing applications.

Purpose of the Study:

• To develop a variable threshold voice activity detector (VAD) for controlling a two-sensor adaptive noise canceller (ANC).
• To enhance ANC performance by preventing speech signal contamination in the reference input during adaptation.
• To reduce the computational complexity of the adaptive filtering process.

Main Methods:

• A novel VAD approach is proposed, utilizing the residual output of the ANC to dynamically adjust thresholds.
• Full-band energy and zero-crossing features are employed and modulated by the noise canceller's residual output.
• The VAD controls the adaptation of the ANC, specifically prohibiting speech during adaptation periods.

Main Results:

• The proposed VAD-controlled ANC demonstrated improved noise cancellation performance across various environmental noise conditions.
• Evaluations showed significant signal-to-noise ratio (SNR) improvements and effective mean square error (MSE) convergence.
• The computational load of the adaptive process was reduced by calculating the adaptive filter output only during non-speech periods.

Conclusions:

• The integration of a residual-output-controlled VAD offers a significant advancement in ANC technology.
• This method effectively enhances noise cancellation efficacy and efficiency.
• The approach provides a computationally lighter and more robust solution for adaptive noise reduction systems.