Description

Created a custom AI model for one of Italy’s leading concert hall lighting technician companies to convert audio tracks into DMX files using pre-recorded lighting plans. The solution analyzes musical structure—such as tempo, intensity, and transitions—and intelligently maps it to synchronized lighting cues, significantly reducing the time required for manual programming. This automation streamlined the setup process for live events and tours, allowing lighting designers to focus on creative adjustments rather than starting from scratch for each performance.

Key Contributions

• Features extraction: Developed an algorithm to extract key features from the audio record, sampled at 20hz.

• AI model: Used a LSTM model to fit the audio features to a standardized DMX file, adjusting the number of channels to the venue and light positioning.

• Iterative appraoch: Followed subsequent iterations to improve the model performance.

• native development: Used Flutter+Bubble to efficiently allow the app to run on the company's tablets and PCs

Technologies Used

• Front-End: Flutter+Bubble

• Back-End: FastAPI and Redis for handling server-side logic and database interactions.

• Database: PostgreSQL for flexible data storage and retrieval.

• AI: Used PyTorch and Polars to extract the sound features and convert them to a DMX matrix

Project highlights

• The first tests cut the preparation time by ~60% for EDM concerts and events.