Multimodal Integration of Mel Spectrograms and Text Transcripts for Enhanced Automatic Speech Recognition: Leveraging Extractive Transformer‐Based Approaches and Late Fusion Strategies