Frequent uroflowmetry testing is inconvenient and contributes to high clinic traffic in low resource environments. Low-cost alternatives for routine home monitoring could alleviate this burden; however, these remain underdeveloped. This study investigates a deep learning approach for predicting urinary flow rate from recorded audio to support comfortable, low-cost home-based assessment. Voiding events were analyzed using a convolutional neural network trained to map acoustic features to continuous flow values. The model successfully captured the temporal pattern of voiding, demonstrating the ability to reconstruct the general shape and progression of the flow curve from sound alone. The model achieved a Mean Absolute Error (MAE) of 2.92 ml/s, Root Mean Squared Error (RMSE) of 3.7 ml/s and R-squared value of 0.74; of its predicted values 68.9% fell within ±3 ml/s of the Minze measured value. However, performance errors were observed at higher flow rates. Despite this limitation, the results demonstrate the feasibility of sound-based uroflowmetry using convolutional neural networks and highlight clear opportunities for improvement, including expanded feature extraction that incorporates additional acoustic features alongside the mel-spectrogram. Overall, this work supports the potential of sound-based monitoring as a practical tool for continuous, at-home evaluation of urinary function.