Wave-Tacotron: Spectrogram-free end-to-end text-to-speech synthesis

TL;DR

Wave-Tacotron introduces a spectrogram-free, end-to-end neural TTS model that directly synthesizes speech waveforms from text, leveraging normalizing flows for parallel training and high-quality output.

Contribution

It extends Tacotron by integrating normalizing flows, enabling direct waveform generation without intermediate features and improving synthesis speed.

Findings

Produces speech quality close to state-of-the-art systems

Allows parallel training and faster synthesis

Eliminates need for intermediate spectrogram representations

Abstract

We describe a sequence-to-sequence neural network which directly generates speech waveforms from text inputs. The architecture extends the Tacotron model by incorporating a normalizing flow into the autoregressive decoder loop. Output waveforms are modeled as a sequence of non-overlapping fixed-length blocks, each one containing hundreds of samples. The interdependencies of waveform samples within each block are modeled using the normalizing flow, enabling parallel training and synthesis. Longer-term dependencies are handled autoregressively by conditioning each flow on preceding blocks.This model can be optimized directly with maximum likelihood, with-out using intermediate, hand-designed features nor additional loss terms. Contemporary state-of-the-art text-to-speech (TTS) systems use a cascade of separately learned models: one (such as Tacotron) which generates intermediate features (such as spectrograms) from text, followed by a vocoder (such as WaveRNN) which generates waveform samples from the intermediate features. The proposed system, in contrast, does not use a fixed intermediate representation, and learns all parameters end-to-end. Experiments show that the proposed model generates speech with quality approaching a state-of-the-art neural TTS system, with significantly improved generation speed.