---
title: 'Singing voice conversion'
---

Singing voice conversion
===

## Table of Contents

[TOC]

## Latest related work
1. 2019 Unsupervised Singing Voice Conversion
- https://arxiv.org/pdf/1904.06590.pdf
- [Nachmani, et al., INTERSPEECH’19]
- Facebook AI Research
- demo: https://enk100.github.io/Unsupervised_Singing_Voice_Conversion/
2. 2020 PITCHNET: Unsupervised Singing Voice Conversion with Pitch Adversarial Network
- https://arxiv.org/pdf/1912.01852.pdf
- [Deng, et al., ICASSP’20]
- Tencent AI Lab
- demo: https://tencent-ailab.github.io/pitch-net/
3. 2020 Unsupervised Cross-Domain Singing Voice Conversion
- https://arxiv.org/pdf/2008.02830.pdf
- Facebook AI Research

## Datasets
1. Standford CCMA - Smule Karaoke App
- https://ccrma.stanford.edu/damp/
- > [name=2019 FB AI Research]
From the “DAMP-multiple” section of this dataset, we selected five singers at random.
Excluding several singers with low quality audio. Each singer has 10 vocal songs, out of which 9 songs are used for training, and the tenth for validation.
2. NUS-48E
- https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6694316
- [downloads here](https://drive.google.com/open?id=12pP9uUl0HTVANU3IPLnumTJiRjPtVUMx)
- 12 singers with 4 songs for each singer
- 48 pairs of sung and spoken
3. LCSING
- https://singing-conversion.github.io
- LJS - voice learned from single speaker speech dataset
- LCSING - voice learned from single speaker singing dataset
- VCTK - Voices learned from multi-speaker speech dataset
- NUS-48E - Voices learned from multi-speaker singing dataset

###### tags: `SingingSynthesis`