This study addresses the critical challenge of enhancing Automatic Speech Recognition (ASR) accuracy for emergency calls in noisy environments by leveraging signal preprocessing techniques. Focusing on audio recordings from Ecuador’s Integrated Security Service (ECU 911), we propose a pipeline combining Low-Pass (LPF) and Band-Pass Filters (BPF) to attenuate ambient noise and isolate relevant speech frequencies. Using OpenAI’s Whisper model for transcription, we evaluate performance and accuracy through Word Error Rate (WER) metrics, comparing the text generated by the model after preprocessing with the reference text. Experimental results demonstrate that cascading LPF and BPF filtering reduces WER in 75% of noisy audios statistically significant reduction, while quiet audios benefit from frequency-selective thresholds. Our findings underscore the efficacy of adaptive filter configurations in improving ASR reliability for emergency response systems, offering a scalable solution for real-world applications.