Samples for "Unsupervised Polyglot Text-to-Speech"

Abstract

We present a Text To Speech Neural network that is able to produce speech in multiple languages. The proposed network is able to transfer a voice, which was presented as a sample in a source language, into one of several target languages. Training is done without using matching or parallel data, i.e., without samples of the same speaker in multiple languages, making the method much more applicable. The conversion is based on learning a polyglot network that has multiple per-language subnetworks and adding loss terms that preserve the speaker's identity in multiple languages. We evaluate the proposed polyglot neural network with three languages with a total of more then 400 speakers and demonstrate convincing conversion capabilities.


Examples for Unsupervised Polyglot Text-to-Speech

Each row represent differnt speaker, each column represent differnt language. All samples was generated by our system. Audio samples in the dialogal are self language synthesis sentences, while others audio samples are language conversion samples.

Example 1

Source\Target English Spanish German
English
Spanish
German

Example 2

Source\Target English Spanish German
English
Spanish
German

Example 3

Source\Target English Spanish German
English
Spanish
German

Example 4

Source\Target English Spanish German
English
Spanish
German

Example 5

Source\Target English Spanish German
English
Spanish
German