An intrusive method for estimating speech intelligibility from noisy and distorted signals

Area of Science:

• Speech processing
• Acoustics
• Signal processing

Background:

• Objective speech intelligibility metrics are crucial for voice communication systems.
• Existing metrics often focus on single noise or distortion types.
• A comprehensive metric for diverse adverse conditions is needed.

Purpose of the Study:

• To propose and evaluate a novel objective metric, the spectrogram orthogonal polynomial measure (SOPM).
• To predict speech intelligibility for normal-hearing listeners under various adverse conditions.
• To assess the metric's performance across different noise and distortion types.

Main Methods:

• Developed the SOPM metric using Krawtchouk moments for feature extraction from spectrograms.
• Evaluated the metric against subjective intelligibility scores from normal-hearing subjects.
• Tested performance under steady-state and fluctuating noise, clipping, phase jitters, time-frequency segregation, and reverberation.

Main Results:

• The SOPM metric demonstrated high correlation (0.97-0.996) with subjective intelligibility scores.
• The metric performed well across a wide range of noise and distortion conditions.
• Achieved robust predictions for speech intelligibility in both quiet and noisy environments.

Conclusions:

• The proposed spectrogram orthogonal polynomial measure (SOPM) is a promising objective metric for predicting speech intelligibility.
• SOPM offers a reliable and accurate method for assessing speech quality under diverse adverse conditions.
• This metric can advance the development of robust voice communication technologies.