Intelligent Harmonization Techniques for Embedded Musical Systems

Abstract

The use of computational systems in musical performance and composition has been increasing in prevalence as the devices involved have grown both smaller and more powerful. Laptops and “stompboxes,” embedded systems typically featuring a digital signal processing (DSP)-capable microprocessor and audio input and output (I/O), are staples of modern-day performances, yet have historically fulfilled vastly different roles on the stage. We propose that a novel device, the Vocodec, is needed to bridge the present divide between signal-chain effects pedals and laptops. Including the necessary audio I/O, physical interface, and amplification, as well as a DSP-capable chip, the Vocodec can stand alone as both a hardware interface and software instrument. To demonstrate the potential of such a device, we propose an application uniquely suited to this platform: intelligent harmonization. Using techniques based on constraint satisfaction algorithms, our harmonization algorithm is able to generate significant compositional complexity with sparse user input, all the while providing physical control of key parameters of the instrument. Given past harmonizations and pitch detection data, the algorithm can produce harmonizations in numerous different styles. For example, the device may perform triadic voice-leading by heuristically minimizing the component-wise distance between successive triads as the user’s MIDI input transitions from note to note, or may analyze the user’s vocal input and provide a responsive single-voice harmonization. Structured and specified as such, the problem is of tractable complexity for real-time computation on an embedded system. The synthesized results are intelligent and intuitive, and complement, rather than compose, the live performance.