WaveNODE: A Continuous Normalizing Flow for Speech Synthesis
This work addresses memory constraints for real-time speech synthesis applications, though it appears incremental as it builds on existing flow-based methods.
The paper tackles the problem of memory inefficiency in flow-based generative models for speech synthesis by proposing WaveNODE, a continuous normalizing flow that eliminates the need for teacher networks or many flow steps, achieving comparable performance with fewer parameters.
In recent years, various flow-based generative models have been proposed to generate high-fidelity waveforms in real-time. However, these models require either a well-trained teacher network or a number of flow steps making them memory-inefficient. In this paper, we propose a novel generative model called WaveNODE which exploits a continuous normalizing flow for speech synthesis. Unlike the conventional models, WaveNODE places no constraint on the function used for flow operation, thus allowing the usage of more flexible and complex functions. Moreover, WaveNODE can be optimized to maximize the likelihood without requiring any teacher network or auxiliary loss terms. We experimentally show that WaveNODE achieves comparable performance with fewer parameters compared to the conventional flow-based vocoders.