In Laser Welding (LW), multiple sources of data can be used to perform process monitoring. Acoustic Emission (AE) has demonstrated advantages since it does not require severe adaptations into the existing system. Optical microphones, specifically, are capable of sampling signals in the order of MHz, opening a vast possibility for monitoring on high frequency domains. In this work, two methodologies of processing AE are presented, assessing the potential of optical microphones as a robust data source for LW and predictor of the laser power. The experimental setup consisted of 22 bead-on-plate runs on E36 steel, with different laser powers, from 1 kW to 6 kW in 500 W intervals. The experiment was monitored via an optical AE microphone at a sampling rate of 2 MHz, and the acquired signals were split in segments of regular intervals. The first methodology is based on the TSFEL library for feature extraction from the data and the usage of Machine Learning (ML) regressors to predict the laser power. The second is based on computing spectrograms using Short Time Fourier Transform (STFT) and a Convolutional Neural Network (CNN) to predict the laser power. Additionally, each datapoint was then transformed via a 2-dimensional Principal Component Analysis (PCA) reduction for qualitative evaluation. Based on a test set evaluation on unseen data, both methods have achieved a strong prediction performance for the laser power, resulting in a R2 of approximately 0.92 and MAE of approximately 0.3kW. The methodology proposed in this work presents an advancement in AE processing, enabling a digital-first, automated LW monitoring system.