A Generative Voice Driven Percussion Application
Author
Vogiatzoglou, Iakovos
Term
4. Term
Education
Publication year
2016
Submitted on
2016-05-27
Abstract
This thesis presents the design and implementation of an interactive, voice-driven percussion application that maps beatboxed vocal patterns to drum samples and can continue or generate new rhythms in the same style as the user’s input. The aim is to leverage the accessibility of the human voice as a musical interface for rapid rhythm sketching and production. The system analyzes vocal percussion, extracts timbral features (including MFCCs), and uses the TimbreID library in Pure Data to train/classify sounds and trigger synthetic drums. For sequence prediction and generation, a Variable Length Markov Model implemented in Python learns from the input, with components communicating via the Open Sound Control (OSC) protocol. The thesis reviews background on the voice as an instrument, prior work, feature extraction, onset detection, and Markov-based models, and describes a prototype spanning PD and Python. The work is currently focused on percussive patterns, with planned extensions to humming and whistling. This excerpt does not report evaluation results; these are discussed later in the thesis.
Denne afhandling præsenterer design og implementering af en interaktiv, stemmestyret percussion-applikation, der omsætter beatbox-mønstre til trommesamples og kan fortsætte eller generere nye rytmer i samme stil som brugerens input. Formålet er at udnytte menneskestemmens tilgængelighed som musikalsk grænseflade til hurtig skitsering og produktion af rytmer. Systemet analyserer vokal percussionsignaler, udtrækker timbrale træk (bl.a. MFCC) og træner/klassificerer lyde med TimbreID i Pure Data for at trigge syntetiske trommer. Til sekvensforudsigelse og -generering anvendes en Variable Length Markov Model implementeret i Python, og komponenterne kommunikerer via Open Sound Control (OSC). Afhandlingen gennemgår relevant baggrund om stemmen som instrument, tidligere arbejde, feature-ekstraktion, onset-detektion og Markov-baserede modeller, samt beskriver en PD- og Python-baseret prototype. Arbejdet fokuserer på percussive mønstre, med planer om at udvide til nynnen og fløjten som fremtidigt arbejde. Dette uddrag rapporterer ikke evalueringsresultater; disse behandles senere i afhandlingen.
[This apstract has been generated with the help of AI directly from the project full text]