ModernBERT-4-Disfluency

Production-ready disfluency detection model for speech processing.

Performance Metrics

• F1 Score: 0.9527
• Precision: 0.9678
• Recall: 0.9380
• Accuracy: 0.9817
• False Positive Rate: 0.0076

Usage

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

tokenizer = AutoTokenizer.from_pretrained("modernBERT-4-disfluency")
model = AutoModelForTokenClassification.from_pretrained("modernBERT-4-disfluency")

detector = pipeline("token-classification", model=model, tokenizer=tokenizer)
results = detector("I um think that uh this works well")

Model Details

• Base Architecture: ModernBERT
• Task: Token Classification for Disfluency Detection
• Training: A100 GPU optimized with advanced loss functions
• Labels: FLUENT (0), DISFLUENT (1)