As part of a research project carried out at the Computer Music and Neural Audio Communication Group and the Audio Communication Group at Technische Universität Berlin, a heterogeneous dataset was created to train an artificial intelligence system that can imitate the sound of an analog plate reverb used in the university’s labs. The goal was to create a digital version of this audio effect, keeping its original sound as faithful as possible. To build the dataset, several input audio signals were collected, chosen to include a good variety of timbres, dynamics, and content. Then, the audio material was analyzed by calculating its main features, using Python scripts with the Librosa library. The final dataset provides a solid base for training the model, helping the AI learn how the plate reverb behaves. This work is part of a larger research effort focused on the digital simulation of analog sound effects using artificial intelligence. Alongside this dataset, a more homogeneous one was also developed to support even more stable and effective training.
Nel contesto di un progetto di ricerca svolto presso il Computer Music and Neural Audio Communication Group e l’Audio Communication Group della Technische Universität Berlin, è stato creato un dataset eterogeneo per allenare un’intelligenza artificiale in grado di imitare il suono di un plate reverb analogico presente nei laboratori dell’ateneo. L’obiettivo era quello di digitalizzare questo effetto audio mantenendo il più possibile le sue caratteristiche originali. Per costruire il dataset sono stati raccolti diversi segnali audio di input, scelti per includere una buona varietà di timbri, dinamiche e contenuti. Successivamente, è stata eseguita l’analisi del materiale audio attraverso il calcolo delle principali feature scrivendo script Python utilizzando la libreria Librosa. Il dataset finale rappresenta una base solida per la fase di training del modello, permettendo all’IA di apprendere in modo efficace il comportamento del riverbero. Questo lavoro si inserisce nella più ampia ricerca sulla simulazione digitale di effetti sonori analogici tramite intelligenza artificiale. Oltre a questo dataset, è stato contemporaneamente sviluppato un dataset omogeneo per ottenere un training ancora più solido ed efficace.
Realizzazione di un heterogeneous dataset per l'emulazione del plate reverb. Uno studio di caso al Computer Music and Neural Audio Communication Group, Audio Communication Group, Technische Universität Berlin.
PAVARIN, PIERPAOLO MARIA
2024/2025
Abstract
As part of a research project carried out at the Computer Music and Neural Audio Communication Group and the Audio Communication Group at Technische Universität Berlin, a heterogeneous dataset was created to train an artificial intelligence system that can imitate the sound of an analog plate reverb used in the university’s labs. The goal was to create a digital version of this audio effect, keeping its original sound as faithful as possible. To build the dataset, several input audio signals were collected, chosen to include a good variety of timbres, dynamics, and content. Then, the audio material was analyzed by calculating its main features, using Python scripts with the Librosa library. The final dataset provides a solid base for training the model, helping the AI learn how the plate reverb behaves. This work is part of a larger research effort focused on the digital simulation of analog sound effects using artificial intelligence. Alongside this dataset, a more homogeneous one was also developed to support even more stable and effective training.| File | Dimensione | Formato | |
|---|---|---|---|
|
Pavarin_PierpaoloMaria.pdf
accesso aperto
Dimensione
4.94 MB
Formato
Adobe PDF
|
4.94 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/92511