Developed audio preprocessing enhancements for the VibeVoice ASR system in the Blaizzy/mlx-audio repository, focusing on stabilizing input quality and improving transcription accuracy. The work introduced audio resampling and loudness normalization, ensuring all inputs are converted to 24 kHz and normalized to -25 dBFS before processing. Sampling_rate-aware APIs were implemented to handle varied input sample rates, and decoding parameters were aligned with the official demo pipeline for consistency. Using Python, audio processing, and machine learning techniques, these changes reduced transcription errors, improved robustness across diverse audio sources, and facilitated easier integration and benchmarking for downstream users and customers.