How can I change the database and pretrained generator ?

4 visualizzazioni (ultimi 30 giorni)
I am working on this example
And I tried to change the database from drums to human voice by changing the track from here
%% Load Training Data
% Train a GAN using the Drum Sound Effects dataset [1].
% Download and extract the dataset.
%%
%url = 'http://deepyeti.ucsd.edu/cdonahue/wavegan/data/drums.tar.gz';
downloadFolder = tempdir;
filename = fullfile(downloadFolder,'drums_dataset.tgz');
drumsFolder = fullfile(downloadFolder,'drums');
But every time I change the track it still plays the sound of the drums!!
A colleague suggested to me changing the pretrained generator!!..
%%
% Download the pretrained generator.
matFileName = 'drumGeneratorWeights.mat';
%if ~exist(matFileName,'file')
% websave(matFileName,'https://www.mathworks.com/supportfiles/audio/GanAudioSynthesis/drumGeneratorWeights.mat');
%end
ِِAnd now I am trying to link the relationship between the database and the pretrained generator..
And how do I change the pretrained generator if I need to change it?
And why is the pretrained generator used at the beginning of the example?
I am a beginner in Matlab .. Any information will help me ... Thank you

Risposte (1)

jibrahim
jibrahim il 31 Gen 2022
Hi Aseel,
The example starts off by showcasing the pretrained network. The network was trained to produce drum-like sounds. If you want it to produce other types of sound, you will need to use another dataset and re-train the network. You can go through the 'Train the GAN' section to see how that is done.
  1 Commento
Sebastian Veron Paterson
Sebastian Veron Paterson il 15 Mag 2024
Hello, if i want to create GAN that can create piano sounds. Do i still need the pre-trained generator and the STFT?

Accedi per commentare.

Categorie

Scopri di più su Audio I/O and Waveform Generation in Help Center e File Exchange

Tag

Prodotti


Release

R2020b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by