Brain-like emergent auditory learning: A developmental method.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Speech Processing

Background:

• The human auditory system excels at accurate and rapid speech recognition, surpassing current machine audition systems.
• Existing artificial auditory models often struggle with complex speech context integration.

Purpose of the Study:

• To propose a new developmental network (DN) that simulates the human auditory pathway for enhanced speech recognition.
• To incorporate a novel layer simulating the superior colliculus for speech context integration.

Main Methods:

• The developmental network (DN) models key elements of the human auditory pathway using deep networks.
• Mel-frequency cepstral coefficients (MFCC) are employed for speech signal feature extraction.
• An additional network layer simulates the superior colliculus's role in thalamic speech context integration.

Main Results:

• The DN model demonstrates emergent features explaining internal neuron representation of short speech context without external supervision.
• Experimental results show significant improvements in English word and phrase recognition rates compared to existing literature.
• The model offers a new approach to address challenges in universal speech recognition within traditional machine audition systems.

Conclusions:

• The proposed developmental network (DN) offers a biologically inspired and effective model for speech recognition.
• This approach provides a potential pathway for advancing machine audition and other computational applications.
• The learning principles may be adaptable to diverse computational contexts beyond speech recognition.