The design and testing of a noise reduction algorithm based on spectral subtraction

Area of Science:

• Speech processing
• Acoustics
• Psychoacoustics

Background:

• Babble noise significantly impacts speech intelligibility.
• Effective noise reduction algorithms are crucial for hearing aid users.
• Existing algorithms may not adequately preserve speech clarity.

Purpose of the Study:

• To design and evaluate three novel amplitude subtraction-based noise reduction algorithms.
• To assess the impact of these algorithms on speech intelligibility in noise.
• To compare performance across normally hearing and hearing-impaired listeners.

Main Methods:

• Developed three noise reduction algorithms using amplitude subtraction.
• Employed a novel synchro method for noise-magnitude spectrum estimation.
• Conducted psychoacoustic tests using monosyllabic words and carrier phrases.
• Evaluated signals with normally hearing and hearing-impaired subjects.
• Applied POGO-fitting spectral shaping for hearing-impaired participants.

Main Results:

• None of the three noise reduction algorithms improved speech intelligibility for any listener group.
• No significant changes in the overall pattern of speech sound confusion were observed.
• Processed signals had long-term power spectra matched to unprocessed signals.

Conclusions:

• The developed amplitude subtraction algorithms, even with a novel synchro method, did not enhance speech intelligibility.
• Further research is needed to develop noise reduction techniques that effectively improve speech clarity for diverse listening conditions and user groups.