Modeling the level-dependent changes of concurrent vowel scores

Area of Science:

• Auditory Neuroscience
• Speech Perception
• Computational Auditory Modeling

Background:

• Fundamental frequency (F0) differences are crucial for separating concurrent speech.
• Previous models suggested auditory-nerve (AN) fiber phase locking to formants and F0s influences vowel identification but lacked direct testing.
• Understanding level-dependent effects on F0 cues is vital for auditory models.

Purpose of the Study:

• To predict concurrent-vowel identification scores using computational AN model temporal responses.
• To validate the role of AN fiber temporal coding in explaining level-dependent F0 benefits.
• To assess an F0-based segregation algorithm's ability to capture human performance across sound levels.

Main Methods:

• Utilized a computational AN model and a modified F0-based segregation algorithm.
• Modeled temporal responses of AN fibers to vowel formants and F0s.
• Compared model predictions with empirical data on concurrent-vowel identification across sound levels and F0 differences.

Main Results:

• The model accurately predicted level-dependent changes in vowel identification scores, with and without F0 differences.
• The model successfully captured identification scores when only one vowel was correct.
• Predicted F0-benefit qualitatively matched empirical F0-benefit across sound levels.

Conclusions:

• Temporal responses of AN fibers to vowel formants and F0s are sufficient to account for identification score variations.
• The computational model validates the importance of AN temporal coding in concurrent-vowel perception.
• This approach provides a quantitative link between neural coding and speech segregation performance across listening conditions.