Crowdstream
Where the crowd becomes the composer
Crowdstream positions itself in the territory where the body becomes an interface and presence is transformed into signal. The work proposes a rereading of the role of the DJ: no longer as a central figure who directs the dance floor, but as a living system, an organism that listens and responds. It is the crowd that sets the tempo, that decides without consciously deciding, that writes the sonic narrative through its own kinetic energy.
AI-Generated Music
Dynamic music generation and remixing based on real-time audience input and environmental cues.
Audience as Composer
Your presence, reactions, and movements shape the music. You're not just listening, you're creating.
Intelligent Sensing
Advanced computer vision analyzes body language, movement, and emotions to understand crowd energy.
What is Crowdstream?
Crowdstream is an interactive music system that generates and mixes sound in real time, driven by signals from the audience. The collective behavior and presence of the crowd directly shape and determine the characteristics of the performance as it unfolds. Crowdstream combines advanced audio processing techniques to extract features, components, and structural elements from a setlist, enabling infinite musical recombinations. This is complemented by a computer vision engine that captures the audience’s movement and dynamics, synthesizing signals that actively drive the composition and performance.
Team Members
Sensing the Audience
Crowdstream reads signals from audience movements and emotions using computer vision techniques, interpreting body language and facial cues to gauge energy and sentiment in real time. Through advanced pose estimation and signal processing, the system translates human presence into musical parameters. These metrics are then routed to the audio engine, where they directly control music mixing, transformation, and generative processes in real time.
Generating Music
AI systems first preprocess and then remix music dynamically based on audience input and environmental cues, creating an adaptive and expressive musical experience. The system responds to the collective energy, building sonic landscapes that emerge from the crowd itself, blurring the boundaries between performer and audience and turning participation into a central compositional force.
System Overview
Presentation of the Crowdstream project at the Deep Learning for Music Information Retrieval Workshop CCRMA, Stanford in August 2025, showcasing the project’s initial concept as well as the system components and the audio and video processing techniques employed.
Demo: Live DJ-Style Stems Mix
A real-time music mixing engine that combines stems from different song sections (intro, verse, chorus), responding to crowd dynamics and creating seamless transitions between musical elements.