Notochord: A Real-Time MIDI Performance Model

Core Concepts

Notochord is a deep probabilistic model for real-time MIDI performance, enabling interactive musical functions with low latency.

Abstract

Notochord designed for real-time music performance. Probabilistic model trained on Lakh MIDI dataset. Allows interventions at sub-event level for diverse musical functions. Applications include generation, harmonization, improvisation. Emphasizes low-latency processing and interpretability. Detailed theory, implementation, and training process provided.

Stats

Notochord responds to inputs with latency below ten milliseconds. Training code, model checkpoints, and interactive examples are open source.

Quotes

"An instrument may begin to feel more like an extension of the body than an external content production device." - Victor Shepardson

Key Insights Distilled From

Notochord

by Victor Shepa... at arxiv.org 03-19-2024

https://arxiv.org/pdf/2403.12000.pdf

Deeper Inquiries

How can Notochord's low-latency capabilities impact live music performances

Notochord's low-latency capabilities can significantly impact live music performances by enabling real-time interaction between the performer and the AI model. With latency below ten milliseconds, Notochord allows for instantaneous responses to user inputs, making it feel like an extension of the performer's body rather than an external device. This seamless integration can enhance the overall experience for both performers and audiences, creating a more immersive and dynamic musical performance.

What challenges might arise when using Notochord in collaborative improvisation settings

When using Notochord in collaborative improvisation settings, several challenges may arise. One key challenge is maintaining coherence and harmony between human musicians and the AI-generated elements. Since Notochord operates based on probabilistic models trained on MIDI data, ensuring that its generated sequences align with the intentions and expressions of human players can be complex. Additionally, coordinating timing, dynamics, and musical motifs between multiple improvising entities—both human and AI—requires careful synchronization to avoid discordance or disruptions in the performance flow.

How does Notochord contribute to the evolving landscape of AI in creative workflows

Notochord contributes to the evolving landscape of AI in creative workflows by offering a versatile tool for interactive music generation in real-time settings. Its deep learning-based probabilistic model allows for interpretability at a sub-event level, enabling diverse interactive musical functions such as steerable generation, harmonization, machine improvisation, and likelihood-based interfaces. By providing open-source software with training code, model checkpoints, and interactive examples, Notochord empowers artists and researchers to explore new possibilities at the intersection of artificial intelligence and music creation. The flexibility of Notochord makes it a valuable asset for studying embodied experiences of AI within musical instruments while pushing boundaries in creative applications leveraging machine learning technologies.