ParrotTTS: Text-to-Speech synthesis by exploiting self-supervised representations
This addresses the problem of data-efficient and flexible TTS synthesis for multilingual applications, though it appears incremental as it builds on existing self-supervised methods.
The paper tackles text-to-speech synthesis by leveraging self-supervised speech representations, enabling effective multi-speaker training with single-speaker transcripts, low-resource language adaptation, and cross-lingual voice transfer without bilingual data, outperforming state-of-the-art models using less paired data.
We present ParrotTTS, a modularized text-to-speech synthesis model leveraging disentangled self-supervised speech representations. It can train a multi-speaker variant effectively using transcripts from a single speaker. ParrotTTS adapts to a new language in low resource setup and generalizes to languages not seen while training the self-supervised backbone. Moreover, without training on bilingual or parallel examples, ParrotTTS can transfer voices across languages while preserving the speaker specific characteristics, e.g., synthesizing fluent Hindi speech using a French speaker's voice and accent. We present extensive results in monolingual and multi-lingual scenarios. ParrotTTS outperforms state-of-the-art multi-lingual TTS models using only a fraction of paired data as latter.