ClariNet: Parallel Wave Generation in End-to-End Text-to-Speech
This work addresses the speed bottleneck in text-to-speech synthesis for real-time applications, offering a novel end-to-end neural architecture that is fully convolutional and enables fast training.
The authors tackled the problem of slow autoregressive WaveNet synthesis by proposing ClariNet, a parallel wave generation method using distillation from an autoregressive WaveNet to a Gaussian inverse autoregressive flow, achieving efficient training and outperforming previous text-to-speech pipelines.
In this work, we propose a new solution for parallel wave generation by WaveNet. In contrast to parallel WaveNet (van den Oord et al., 2018), we distill a Gaussian inverse autoregressive flow from the autoregressive WaveNet by minimizing a regularized KL divergence between their highly-peaked output distributions. Our method computes the KL divergence in closed-form, which simplifies the training algorithm and provides very efficient distillation. In addition, we introduce the first text-to-wave neural architecture for speech synthesis, which is fully convolutional and enables fast end-to-end training from scratch. It significantly outperforms the previous pipeline that connects a text-to-spectrogram model to a separately trained WaveNet (Ping et al., 2018). We also successfully distill a parallel waveform synthesizer conditioned on the hidden representation in this end-to-end model.