A cross-spectrum weighting algorithm for speech enhancement and array processing: combining phase-shift information...

Area of Science:

• Acoustics and Signal Processing
• Speech Enhancement
• Array Signal Processing

Background:

• Microphone arrays are crucial for noise cancellation in complex acoustic environments like vehicles.
• Existing single- and multi-channel algorithms have limitations in accurately estimating speech signals amidst noise.
• Robust speech signal estimation is essential for effective noise reduction and improved communication.

Purpose of the Study:

• To present a novel gain function for noise cancellation using a two-channel microphone array.
• To combine the strengths of single- and multi-channel noise cancellation algorithms.
• To develop an extended gain function that accounts for speech pauses and spectral variations.

Main Methods:

• Developed a gain function based on a minimum mean square error (MMSE) estimator.
• Utilized the cross-spectrum between two microphone signals for speech amplitude estimation.
• Extended the gain function to handle non-stationary speech characteristics, including pauses.

Main Results:

• The proposed gain function effectively cancels diffuse driving noise in a car.
• Significant cancellation of interfering speech signals was achieved using the two-channel microphone array.
• The extended gain function demonstrated improved performance by considering speech pauses.

Conclusions:

• The novel gain function offers a robust solution for noise cancellation in challenging acoustic environments.
• Combining MMSE estimation with cross-spectral analysis provides accurate speech amplitude estimation.
• The extended gain function enhances noise reduction performance by adapting to speech signal dynamics.