FloWaveNet : A Generative Flow for Raw Audio
Sungwon Kim, Sang-gil Lee, Jongyoon Song, Jaehyeon Kim, Sungroh Yoon
TL;DR
FloWaveNet presents a flow-based, conditional generative framework for raw audio that enables real-time, parallel synthesis with a single maximum-likelihood loss and end-to-end training. By stacking multiple affine coupling flows inside context blocks and employing ActNorm, FloWaveNet achieves high-quality audio comparable to two-stage parallel models while avoiding teacher networks and auxiliary losses. The approach demonstrates real-time sampling at typical speech rates and offers a practical drop-in replacement for WaveNet vocoders in various TTS pipelines. The work includes open-source references and a comparative study against Gaussian IAF, highlighting the method's stability, efficiency, and perceptual performance gains in real-world speech synthesis tasks.
Abstract
Most modern text-to-speech architectures use a WaveNet vocoder for synthesizing high-fidelity waveform audio, but there have been limitations, such as high inference time, in its practical application due to its ancestral sampling scheme. The recently suggested Parallel WaveNet and ClariNet have achieved real-time audio synthesis capability by incorporating inverse autoregressive flow for parallel sampling. However, these approaches require a two-stage training pipeline with a well-trained teacher network and can only produce natural sound by using probability distillation along with auxiliary loss terms. We propose FloWaveNet, a flow-based generative model for raw audio synthesis. FloWaveNet requires only a single-stage training procedure and a single maximum likelihood loss, without any additional auxiliary terms, and it is inherently parallel due to the characteristics of generative flow. The model can efficiently sample raw audio in real-time, with clarity comparable to previous two-stage parallel models. The code and samples for all models, including our FloWaveNet, are publicly available.