The infamous data science workflow with interconnected circles of data acquisition, wrangling, analysis, and reporting understates the multi-connectivity and non-linearity of these components. The same is true for machine learning and deep learning workflows. I understand the need for oversimplification is expedient in presentations and executive summaries. However, it may paint unrealistic pictures, hide the intricacies of ML development and conceal the realities of the mess.

Audio signals are all around us. As such, there is an increasing interest in audio classification for various scenarios, from fire alarm detection for hearing impaired people, through engine sound analysis for maintenance purposes, to baby monitoring. Though audio signals are temporal in nature, in many cases it is possible to leverage recent advancements in the field of image classification and use popular high performing convolutional neural networks for audio classification.

The design and training of neural networks are still challenging and unpredictable procedures. The difficulty of tuning these models makes training and reproducing more of an art than a science, based on the researcher’s knowledge and experience. One of the reasons for this difficulty is that the training procedure of machine learning models includes multiple hyperparameters that affect how the training process fits the model to the data.

So, if you’re a nose-to-the-keyboard developer, there’s ample probability that this analogy is outside your comfort zone … bear with me. Imagine two Olympics-level figure skaters working together on the ice, day in and day out, to develop and perfect a medal-winning performance. Each has his or her role, and they work in sync to merge their actions and fine-tune the results.