Von: Marc Voigt via ak discourse
Datum: Fri, 24 May 2024
Betreff: [ak-discourse] Einladung zum Forschungskolloquium am 28.05.2024
Liebe Freunde der Audiokommunikation,
am kommenden Dienstag, den 28.05.2024, um 16:00 Uhr präsentiert Pablo Combes seine Masterarbeit zum Thema
„Differentiable Proxy of Sound Synthesizers: Learning Perceptually informed Presets Representation“
im Raum EN 324.
Dazu möchten wir Sie herzlich einladen. Eine Kurzzusammenfassung finden Sie am Ende dieser E-Mail.
Der Zoom-Link für auswärtige Gäste lautet:
https://tu-berlin.zoom.us/j/64026853839?pwd=VzZWUU1ldjZlem9ZdlUyNUNFaHZxUT09
Meeting-ID: 640 2685 3839
Kenncode: 704269
Mit herzlichen Grüßen
Marc Voigt & Stefan Weinzierl
—
Pablo Combes: Differentiable Proxy of Sound Synthesizers: Learning Perceptually informed Presets Representation
This master’s thesis presents a simple method for approximating black-box sound synthesizers. It is primarily motivated by the need to address the limitations of parameter loss functions used in neural-based methods for Automatic Synthesizer Programming (nASP) in handling the highly nonlinear relationship between synthesizer parameters and synthesized audio. The proposed method relies on training a neural network capable of mapping synthesizer presets onto a perceptually informed embedding space defined by a pretrained audio model. This process effectively creates a neural proxy for a given synthesizer by leveraging the audio representations learned by the pretrained model through a cross-modal knowledge distillation task. The effectiveness of various neural network architectures, including feedforward, recurrent, and transformer-based models, in encoding synthesizer presets was evaluated using both synthetic and hand-crafted presets from three popular software synthesizers. Encouraging results were obtained for all synthesizers, paving the way for future research into the application of synthesizer proxies for nASP methods, focusing on non-differentiable, black-box synthesizers.
—
Marc Voigt
IT-Administration
Technische Universität Berlin
Fakultät I – Geistes- und Bildungswissenschaften
Institut für Sprache und Kommunikation
Fachgebiet Audiokommunikation
Faculty I – Humanities and Educational Sciences
Institute of Speech and Communication
Audio Communication Group
Einsteinufer 17c, 10587 Berlin
GERMANY
Telefon: +49 (0)30 314-25557
Telefax: +49 (0)30 314-21143
marc.voigt@tu-berlin.de