Voice Recognition Performance Improvement using Supplemented SFN for AI Voice Learning Efficiency

Key Points

• The aim is to enhance voice recognition performance using Silence Feature Normalization (SFN) in artificial intelligence models.
• Utilized improved Silence Feature Normalization (SFN) for training AI voice recognition models.
• Evaluated voice recognition accuracy based on Signal-to-Noise Ratio (SNR) thresholds.
• Analyzed Noise Reduction Rate and Pause Hit Ratio (PHR) to assess performance.
• Achieved higher voice recognition accuracy in high SNR conditions with minimal noise impact.
• Enhanced recognition rate observed using feature distribution characteristics at low SNR levels.
• Demonstrated significant improvements in recognition accuracy, especially in noisy settings.

Abstract