# AUTOVC: Zero-Shot Voice Style Transfer with Only Autoencoder Loss

Source: [AUTOVC: Zero-Shot Voice Style Transfer with Only Autoencoder Loss](<https://www.sota2.com/research/paper/68fa65d81e209131cea0cb78>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-25T21:36:37.771Z

Authors: Kaizhi Qian, Yang Zhang, Shiyu Chang, Xuesong Yang, Mark Hasegawa-Johnson

Published: 2019-05-14T23:19:04.000Z

## Abstract

Non-parallel many-to-many voice conversion, as well as zero-shot voice conversion, remain under-explored areas. Deep style transfer algorithms, such as generative adversarial networks (GAN) and conditional variational autoencoder (CVAE), are being applied as new solutions in this field. However, GAN training is sophisticated and difficult, and there is no strong evidence that its generated speech is of good perceptual quality. On the other hand, CVAE training is simple but does not come with the distribution-matching property of a GAN. In this paper, we propose a new style transfer scheme that involves only an autoencoder with a carefully designed bottleneck. We formally show that this scheme can achieve distribution-matching style transfer by training only on a self-reconstruction loss. Based on this scheme, we proposed AUTOVC, which achieves state-of-the-art results in many-to-many voice conversion with non-parallel data, and which is the first to perform zero-shot voice conversion.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Voice Conversion on VCTK seen-to-seen (test)](<https://www.sota2.com/research/sota/voice-conversion-on-vctk-seen-to-seen-test.md>): Voice Conversion; VCTK seen-to-seen (test); MOS: 3.41.

- [Voice Conversion on LibriTTS to VCTK (unseen-to-seen) (test-clean)](<https://www.sota2.com/research/sota/voice-conversion-on-libritts-to-vctk-unseen-to-seen-test-clean.md>): Voice Conversion; LibriTTS to VCTK (unseen-to-seen) (test-clean); MOS: 2.74.

- [Voice Conversion on LibriTTS unseen-to-unseen (test-clean)](<https://www.sota2.com/research/sota/voice-conversion-on-libritts-unseen-to-unseen-test-clean.md>): Voice Conversion; LibriTTS unseen-to-unseen (test-clean); MOS: 2.59.

- [Voice Conversion on ESD](<https://www.sota2.com/research/sota/voice-conversion-on-esd.md>): Voice Conversion; ESD; WER: 0.87.

- [Voice Conversion on VCTK](<https://www.sota2.com/research/sota/voice-conversion-on-vctk.md>): Voice Conversion; VCTK; WER: 71.3.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/1905.05879>)

[PDF](<https://arxiv.org/pdf/1905.05879.pdf>)
