Learning to Speak Fluently in a Foreign Language: Multilingual Speech Synthesis and Cross-Language Voice Cloning

TL;DR

This paper introduces a multilingual speech synthesis model capable of high-quality, cross-language voice transfer without bilingual data, using phonemic inputs and adversarial training to disentangle speaker identity from language content.

Contribution

The novel model enables cross-language voice cloning and fluent speech synthesis across distant languages without bilingual training data, advancing multilingual TTS technology.

Findings

Effective voice transfer across languages, including distant ones like English and Mandarin.

High-quality speech synthesis in multiple languages with consistent speaker identity.

Model can generate speech in native or foreign accents for multiple speakers.

Abstract

We present a multispeaker, multilingual text-to-speech (TTS) synthesis model based on Tacotron that is able to produce high quality speech in multiple languages. Moreover, the model is able to transfer voices across languages, e.g. synthesize fluent Spanish speech using an English speaker's voice, without training on any bilingual or parallel examples. Such transfer works across distantly related languages, e.g. English and Mandarin. Critical to achieving this result are: 1. using a phonemic input representation to encourage sharing of model capacity across languages, and 2. incorporating an adversarial loss term to encourage the model to disentangle its representation of speaker identity (which is perfectly correlated with language in the training data) from the speech content. Further scaling up the model by training on multiple speakers of each language, and incorporating an autoencoding input to help stabilize attention during training, results in a model which can be used to consistently synthesize intelligible speech for training speakers in all languages seen during training, and in native or foreign accents.