Comparison of Two Speech Processing Schemes Using Normal-hearing Subjects

Area of Science:

• Audiology and Speech Sciences
• Signal Processing
• Psychoacoustics

Background:

• Speech processing schemes are crucial for understanding speech perception, particularly for individuals with hearing impairments.
• Real-time laboratory speech processors enable controlled studies on how spectral information affects auditory perception.
• Comparing different spectral component extraction methods is essential for optimizing speech intelligibility.

Purpose of the Study:

• To compare the efficacy of two distinct speech processing schemes, zero-crossing (ZC) and Filter-bank (FB), in a real-time laboratory setting.
• To evaluate the impact of the number of spectral components (two vs. four) on speech perception in normal-hearing individuals.
• To assess performance differences in vowel and consonant perception, as well as word recognition and speech tracking.

Main Methods:

• Implemented two real-time speech processing schemes: ZC (two spectral components) and FB (four spectral components).
• Conducted perceptual studies with 4 normal-hearing subjects, employing a within-subjects design with counterbalanced training and testing.
• Measured vowel perception, consonant perception in noise (5-20 dB SNR), open-set CNC word recognition, and speech tracking scores.

Main Results:

• Vowel perception performance was comparable between the ZC and FB schemes across all subjects.
• Consonant perception in noise showed significantly higher accuracy for the FB scheme compared to the ZC scheme.
• Open-set CNC word recognition and speech tracking scores were also superior with the FB scheme.

Conclusions:

• The Filter-bank (FB) scheme, utilizing more spectral information (four components), offers enhanced consonant perception in noise over the zero-crossing (ZC) scheme (two components).
• The FB scheme demonstrates superior performance in complex speech recognition tasks like CNC words and speech tracking.
• These findings suggest that processing more spectral detail is beneficial for speech intelligibility, especially in noisy environments.