Speech Recognition Models in Assisting Medical History

This paper addresses challenges highlighted by health professionals, where up to 50\% of a medical consultation's time is spent on history creation. To streamline this process, we propose leveraging Automatic Speech Recognition (ASR) models to convert spoken language into text. In our study, we assess the effectiveness of pre-trained ASR models for medical history transcription in Brazilian Portuguese. By incorporating language models to enhance ASR output, we aim to improve the accuracy and semantic fidelity of transcriptions. Our results demonstrate that integrating a 5-gram model with Wav2Vec2 PT significantly reduces transcription errors, while also maintaining superior performance in capturing textual nuances and similarity.