https://github.com/radoslawregula/voxg

Singing voice synthesizer using GANs
https://github.com/radoslawregula/voxg

audio audio-generation audio-processing deep-learning gan generative-adversarial-network machine-learning music-programming music-technology python singing-voice-synthesis tensorflow voice-synthesis

Last synced: 15 days ago
JSON representation

Singing voice synthesizer using GANs

• Host: GitHub
• URL: https://github.com/radoslawregula/voxg
• Owner: radoslawregula
• License: mit
• Created: 2021-11-13T21:40:12.000Z (about 3 years ago)
• Default Branch: master
• Last Pushed: 2023-03-01T21:58:44.000Z (almost 2 years ago)
• Last Synced: 2024-11-20T07:18:29.189Z (3 months ago)
• Topics: audio, audio-generation, audio-processing, deep-learning, gan, generative-adversarial-network, machine-learning, music-programming, music-technology, python, singing-voice-synthesis, tensorflow, voice-synthesis
• Language: Python
• Homepage:
• Size: 145 KB
• Stars: 2
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
VoxG: Singing voice synthesizer using GAN.

==============================

This repository contains implementation of a Wasserstein-GAN for singing voice synthesis in Tensorflow 2.5. The neural networks were trained using publicly available [NUS-48E sung and spoken lyrics corpus](https://www.smcnus.org/wp-content/uploads/2013/09/05-Pub-NUS-48E.pdf). 

The code is based on an original implementation described in [WGANSing: A Multi-Voice Singing Voice Synthesizer Based on the Wasserstein-GAN](https://arxiv.org/pdf/1903.10729.pdf), however with improvements regarding the training process and code quality. The implementation of WGAN can be reused for other similar tasks.

This project was completed as a Master's thesis in Computer Science, Adam Mickiewicz University, Poznań, Poland.