Sources of Variability in Consonant Perception and Implications for Speech Perception Modeling

Area of Science:

• Auditory Perception
• Speech Processing
• Psychoacoustics

Background:

• Response variability in consonant perception arises from both external (source-induced) and internal (receiver-related) factors.
• Understanding these variability sources is crucial for developing accurate speech perception models.

Purpose of the Study:

• To investigate the influence of various sources of variability on consonant perception.
• To compare the predictive power of energy-based and modulation-based auditory models.

Main Methods:

• Consonant-vowel combinations were presented to normal-hearing listeners in white noise across varying signal-to-noise ratios.
• Perceptual distance analysis quantified variability effects from different sources.
• Two auditory-inspired models (energy-based and modulation-based) were evaluated against perceptual data.

Main Results:

• Variability across different consonant-vowel combinations had the largest effect.
• Speech-induced variability (talker, noise) and across-listener variability were substantial and comparable.
• Modulation-based models demonstrated superior agreement with consonant recognition and confusion data compared to energy-based models.

Conclusions:

• Low-frequency modulations, up to 16 Hz, are critical for consonant perception.
• Modulation-based auditory models provide a better account of consonant recognition and confusions than energy-based models.