Speech-in-noise enhancement using amplification and dynamic range compression controlled by the speech...

Area of Science:

• Speech processing
• Acoustics
• Signal processing

Background:

• Additive noise degrades speech intelligibility in communication systems.
• Existing algorithms struggle with non-stationary noise conditions.
• Speech Intelligibility Index (SII) is crucial for assessing intelligibility.

Purpose of the Study:

• To propose and evaluate the AdaptDRC algorithm for enhanced speech intelligibility.
• To compare AdaptDRC against a modified existing algorithm.
• To assess performance under various noise conditions using objective and subjective measures.

Main Methods:

• Developed AdaptDRC with amplification and dynamic range compression stages, dependent on Speech Intelligibility Index (SII).
• Conducted experiments using extended SII and short-time objective intelligibility (STOI) measures.
• Performed formal listening tests in stationary (car, speech-shaped) and non-stationary (cafeteria) noise.

Main Results:

• Objective measures (extended SII, STOI) showed similar performance between AdaptDRC and the compared algorithm.
• Formal listening tests revealed significant intelligibility improvements for both algorithms in stationary noise.
• AdaptDRC demonstrated statistically significant improvements in non-stationary cafeteria noise, unlike the other algorithm.

Conclusions:

• AdaptDRC effectively enhances speech intelligibility, particularly in challenging non-stationary noise environments.
• Formal listening tests correlate well with objective intelligibility measures.
• Objective measures may overestimate the actual performance of speech enhancement algorithms.