Deep Voice 3: Scaling Text-to-Speech with Convolutional Sequence Learning

TL;DR

Deep Voice 3 is a convolutional neural TTS system that achieves high naturalness, trains rapidly on large multi-speaker datasets, and scales inference efficiently, advancing the state-of-the-art in speech synthesis.

Contribution

It introduces a fully convolutional attention-based TTS model that trains faster, scales to large datasets, and demonstrates scalable inference on a single GPU.

Findings

Matches state-of-the-art naturalness in speech synthesis

Trains ten times faster than comparable systems

Scales inference to ten million queries per day

Abstract

We present Deep Voice 3, a fully-convolutional attention-based neural text-to-speech (TTS) system. Deep Voice 3 matches state-of-the-art neural speech synthesis systems in naturalness while training ten times faster. We scale Deep Voice 3 to data set sizes unprecedented for TTS, training on more than eight hundred hours of audio from over two thousand speakers. In addition, we identify common error modes of attention-based speech synthesis networks, demonstrate how to mitigate them, and compare several different waveform synthesis methods. We also describe how to scale inference to ten million queries per day on one single-GPU server.