Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis

TL;DR

This paper presents a neural TTS system that leverages transfer learning from speaker verification to synthesize natural speech in many voices, including unseen speakers, by using a speaker encoder trained on a large dataset.

Contribution

It introduces a novel transfer learning approach that uses a speaker verification network to improve multispeaker TTS, enabling synthesis of voices not seen during training.

Findings

The system can synthesize speech in unseen speakers' voices.

Training the speaker encoder on a large diverse dataset improves generalization.

Randomly sampled embeddings produce high-quality novel speaker voices.

Abstract

We describe a neural network-based system for text-to-speech (TTS) synthesis that is able to generate speech audio in the voice of many different speakers, including those unseen during training. Our system consists of three independently trained components: (1) a speaker encoder network, trained on a speaker verification task using an independent dataset of noisy speech from thousands of speakers without transcripts, to generate a fixed-dimensional embedding vector from seconds of reference speech from a target speaker; (2) a sequence-to-sequence synthesis network based on Tacotron 2, which generates a mel spectrogram from text, conditioned on the speaker embedding; (3) an auto-regressive WaveNet-based vocoder that converts the mel spectrogram into a sequence of time domain waveform samples. We demonstrate that the proposed model is able to transfer the knowledge of speaker variability learned by the discriminatively-trained speaker encoder to the new task, and is able to synthesize natural speech from speakers that were not seen during training. We quantify the importance of training the speaker encoder on a large and diverse speaker set in order to obtain the best generalization performance. Finally, we show that randomly sampled speaker embeddings can be used to synthesize speech in the voice of novel speakers dissimilar from those used in training, indicating that the model has learned a high quality speaker representation.